Federal Reserve Bank of Dallas

Globalization and Monetary Policy Institute

Working Paper No. 293
https://dallasfedcm.ws.frb.org/-/media/Documents/institute/wpapers/2016/0293.pdf

Financial Regulations and Price Inconsistencies across Bitcoin

Markets*

Gina Pieters
Trinity University

Sofia Vivanco
Trinity University

December 2016

Abstract
We document systematic differences in bitcoin prices across 11 different markets
representing 26% of global bitcoin trade volume. These differences must due to the
identical nature of all bitcoin result from characteristics of markets themselves. We
examine differences across the markets and find that those which do not require customer
identification for establishing an account are more likely to deviate from representative
market prices than those which do. This implies that standard financial regulations,
specifically know-your-customer regulations, can have a non-negligible impact on the bitcoin
market.

JEL codes: G12, G14, K42, L86

*
Gina Pieters, Trinity University, Economics Department, 1 Trinity Place, San Antonio, TX 78212.
gpieters@trinity.edu. Sofia Vivanco, Economics Department, 1 Trinity Place, San Antonio, TX 78212.
This paper was written while Vivanco was an undergraduate at Trinity University. The views in this paper
are those of the authors and do not necessarily reflect the views of the Federal Reserve Bank of Dallas or
the Federal Reserve System.
1 Introduction

Bitcoin is a homogeneous, virtual good that is completely identical across all the online markets in which it

is sold.1 The markets that sell bitcoin (bitcoin exchanges) operate 24 hours a day, 7 days a week, with

pricing data available immediately and free of charge to anyone in the world with Internet access. Despite

these properties, we find that bitcoin prices behave differently across markets; not all exchanges adhere to

the law of one price (LOOP) the theory that identical goods should sell for identical prices after trade

costs are taken into consideration.

Should LOOP be expected of bitcoin prices? Whether bitcoin is considered a currency trading on

international exchange rate markets or as argued by Yermack (2015) an asset trading on international

financial markets, the answer is yes. Akram, Rime, and Sarno (2008) established the existence of LOOP in

exchange rate markets, while Eun and Sabherwal (2003) established LOOP for internationally traded assets.

Thus, if a bitcoin exchange fails to achieve LOOP, the cause must be associated with a characteristic of the

exchange. We find bitcoin markets that consistently fail LOOP do not require customer identification to

initiate an account, and therefore do not fully implement know-your-customer (KYC) policies. We are not

aware of any other research that shows an impact by traditional financial regulations on bitcoin markets.

While there is no uniform, global regulatory framework for bitcoincountries regulate or ignore bitcoin as

they choose most bitcoin regulations were implemented during 2013 or the beginning of 2014. The nature

and extent of these regulations vary with some, such as Kaplanov (2012), debating the need for any regulation,

envisioning a currency free of government intervention. The most recent bitcoin regulation announcement at

the time of writing occurred on July 2014: Ecuador banned bitcoin, while the Isle of Mann merely clarified

the application of existing tax rules.2 Within the USA, virtual currency exchanges are regulated by the

Financial Crimes Enforcement Network (FinCEN), though individual states (e.g. New York) may adopt

tougher measures. In March 2013, FinCEN classified bitcoin markets as be money transmitting businesses

(MTBs), requiring exchanges to adopt anti-money-laundering (AML) and KYC policies. However, the

global nature of bitcoin allows an exchange that trades bitcoin for US dollars to claim that it does not do

business with the US consumers, and is therefore not subject to FinCENs rulings. It is also feasible for an

exchange to only selectively enforce a stated AML/KYC policy. King (2016) proposes a new framework to

encourage financial institutions in the USA to incorporate bitcoin-based businesses (BBBs). A key element

of this framework is the expectation that financial institutions ensure that the BBBs implement and enforce

AML/KYC. Even though the global bitcoin regulatory environment is fractured, results from this paper
1 Whetherbitcoin is a currency, a good, or an asset is not relevant for this paper.
2Adetailed discussion of the global regulatory environment can be found at http://www.coindesk.com/bitcoin-legal-map/
and http://www.coindesk.com/information/is-bitcoin-legal/

2
imply that these policies may have an effect on the bitcoin market.

We provide a brief introduction to relevant bitcoin terminology and price research in Section 2. In Section

3 we summarize the behavior of bitcoin prices in US dollars across twelve price sources, and document the

difference in levels both within and across exchanges depending upon which price-series is used. In Section

4, we document the failure of LOOP for several exchanges in our data. In Section 5 we examine the

characteristics of each exchange, and show that all exchanges that fail LOOP do not require consumers to

provide identification to establish an account. We find lower price deviations on exchanges that require

identification, providing an explanation for these exchanges are more prone to LOOP failure. Section 6

concludes.

2 A Brief Review Of Bitcoin

This section provides the information needed to understand bitcoin for this paper; for a comprehensive

overview of bitcoin, readers should consult Velde (2013). An accessible technical review of bitcoin can be

found in Bohme, Christin, Edelman, and Moore (2015). Brandvold, Molnar, Vagstad, and Valstad (2015)

contains a brief discussion of major events in bitcoin history, while White (2015) considers the market for

crypto-currencies more broadly.

Bitcoin is a digital currency with no central issuing authority or physical form.3 More specifically, bitcoin

is a cryptocurrency, a subset of digital currencies that use cryptography techniques to guide encryption

protocols which identify ownership and verify transactions details. The bitcoin encryption protocol was

outlined by the entity Nakamoto (2008)4 and uniquely identifies each bitcoin within the bitcoin economy.

Nakamoto created 21 million bitcoin, but did not release these into circulation. Instead, each bitcoin must

be discovered by solving mathematical algorithms in a process called mining, examined by Kroll, Davey,

and Felten (2013). Once mined, every bitcoinor fraction thereofcan be sold, used as payment for retail

purchases, or kept as an investment to be traded later. Bitcoin is bought and sold on trading websites known

as exchanges, each of which is independently operated and accessible 24 hours a day, 7 days a week to a

global clientele.

It may seem strange that a completely virtual good could sustain a positive price. Dwyer (2015) showed

that this phenomenon results from the limited availability and unique identification of each bitcoin. Luther

(2016) shows that a requirement for this result is low transaction costs. Halaburda and Gandal (2014)

show that, in its early establishment phase, the value of a cryptocurrency (of which bitcoin is but one
3 Perspectives on whether bitcoin satisfies the economic definition of money can be found in Velde (2013), Lo and Wang

(2014), and Yermack (2015), but do not impact the results of this paper.
4 No one knows whether Nakamoto is an individual or a group, male or female, or actually Japanese.

3
example) is determined by competition with other cryptocurrencies. This creates volatility in its initial

price, but later transitions into pricing behavior consistent with use as a financial asset. This mechanism

explains why Kristoufek (2013) finds a cointegrating relationship between bitcoin values and search queries

on Google Trends and Wikipedia (potentially indicating a speculative bubble driven by curiosity), while

Glaser, Zimmermann, Haferkorn, Weber, and Siering (2014) find that consumers behavior is consistent with

use of bitcoin as an investment vehicle. For investment purposes, Briere, Oosterlinck, and Szafarz (2015)

show that bitcoin is an effective diversification asset.

Bitcoin has gained links to the physical goods economy over time as retailersincluding Windows, Dell,

Overstock, and gift card merchantsbegan accepting it as a form of payment, though not all use the same

method.5 For example, Dell accepts direct bitcoin payments, but Amazon instead provides digital gift cards

which may be purchased with bitcoin and then used to purchase goods on their website. Although a growing

number of vendors accept bitcoin, both Ali, Barrdear, Clews, and Southgate (2014) and Greene and Shy

(2014) conclude that digital currencies such as bitcoin do not currently have enough users to act as a general

purchasing alternative to fiat currencies.

3 About Bitcoin Prices

3.1 Bitcoin Data Collection

We obtain the daily data for all exchanges that post prices in US dollars for a minimum 410 of the 426

days between 1 June 2014, and 31 July 2015, from www.bitcoincharts.com.6 This yields data from eleven

exchanges7 (Table 1), with price and volume statistics summarized in Table 2 and detailed exchange charac-

teristics discussed in Section 5.1. We also collect data from Coindesks Bitcoin Price Index, which aggregates

prices from selected exchanges into a reference price-series. Linear interpolation is used in case of missing

data-values.8

The data consist of the daily trade volume for each exchange, along with the Open, Close, High, Low,

and Transaction-weighted prices. The BPI does not report Transaction-weighted price. In total we consider

59 price-series: five variants from each of the 11 exchanges, and four variants from the BPI. We examine all

five variants to verify that the finding of this paper is a consistent feature of an exchanges prices. Bitcoin
5 Websites such as coinmap (http://coinmap.org) or usebitcoins (usebitcoins.info) maintain lists of businessesboth online

and physicalthat accept bitcoin.

6 All the studied exchanges have their bitcoin prices posted on the same aggregation website, and can therefore be simulta-

neously observed by all participants.

7 One additional exchange, Cointrader, satisfies this selection criterion but is not included as it subsequently declared

bankruptcy due to a hack.

8 The price-series used as reference (BitFinex) and the exchanges with the most troubled price-series (LocalBitcoins and

HitBTC), have no missing data values so linear interpolation is not driving the results of this paper.

4
 Table 1: Selected Bitcoin Exchanges

Observations
Name Initial Interpolated Website
ANXBTC 419 426 https://anxbtc.com
BitFinex 426 426 https://www.bitfinex.com
BitKonan 423 426 https://bitkonan.com
BitStamp 423 426 https://www.bitstamp.net
BTC-e 415 426 https://btc-e.com
HitBTC 426 426 https://hitbtc.com
itBit 426 426 https://www.itbit.com
Kraken 417 426 https://www.kraken.com
LakeBTC 414 414 https://www.lakebtc.com
LocalBitcoins 426 426 https://localbitcoins.com
The Rock Trading 420 426 https://www.therocktrading.com
Bitcoin Price Index (BPI) 426 426 http://www.coindesk.com/price/

exchanges operate 24 hours a day, 7 days a week, the close is the price of the last trade prior to 23:59:59

UTC, open is the price of the first trade after 00:00:00 UTC. We obtain data on the daily total global volume

of bitcoin trades from https://blockchain.info. The BPI does not report a trade volume associated with its

index price.

The data contain several clear outliers. For example, on 3 November 2014 the exchange LakeBTC reports

a High price of $1000 for one bitcoin. The highest LakeBTC price on the day preceding this transaction was

$330.52 and the highest price on August 2nd, the day following, was $329.67. The average High price across

all the other exchanges on November 3rd was $392.88. Outliers could contribute towards a finding of LOOP

failure if kept, and given that determining whether outliers are bad data or an actual transaction price

is beyond the scope of this paperwe remove them to minimize potential false identification. Outliers are

replaced by the average of the adjacent prices on the exchange (in the example above, $1000 is replaced by

$330.10).9 Removing outliers with this method reduces the number of price deviations, and therefore biases

the data against our main finding of non-LOOP behavior.

3.2 Summary of Bitcoin Prices

The Open price-series for each exchange is shown in Figure 1. Visually, the prices for all exchanges are similar,

decreasing from a high of $600, at the beginning of the sample, to a mid-sample low of approximately $200,

followed by a slow increase. The turning point does not correspond to a known event in bitcoin history, and

the exact timing varies across exchanges. The LocalBitcoins price-series are clearly more volatile than other

exchanges, but despite this follow the same trend as other exchanges. LocalBitcoins Transaction-weighted
9 Significant outliers are defined as a price satisfying at least one of the following criteria: A price less than $1 or greater

than $3,000; a percent change in daily price further than 10 standard deviations away from the mean of the changes for the
series; or a price that doubles (or halves) then approximately returns to its original value the following day.

5
 Figure 1: Open USD Bitcoin Price

price displays a volatility comparable to other exchanges.10

The median daily Open, Close, High, Low, and Transaction-weighted price behavior across exchanges is

summarized in Table 2. LocalBitcoins posts a median Open, Close, Weighted, and High price well above

the other exchanges, while Low prices are below those of other exchanges. LocalBitcoins median daily

High/Low Price Premium (the difference between the High and Low price, relative to the Low price) is

136.75%, easily the highest of all the exchanges (the second-highest is BitKonan, where the median daily

High price is 6.12% above the Low price).11 It should come as no surprise that LocalBitcoins data will be

found to be unrepresentative of bitcoin prices. It is less obvious from Table 2 that this distinction is shared

by HitBTC as well. A comparison of bitcoin price levels is therefore not sufficient to identify exchanges that

are not representative of the bitcoin market.

The remaining four columns of Table 2 compare the transaction volumes on each exchange relative to the

estimated transaction volume across the entire bitcoin market, provided by Blockchain (www.blockchain.info,
10 Seriesvolatility is examined in Section A.1.
11 Recall that outliers within each series were removed. Despite this, LocalBitcoins High prices are persistently higher, and
its low prices persistently lower than other exchanges.

6
 Table 2: Median Price and Volume Characteristics Across Exchanges

Price (USD) High/Low Transaction Volume

Open Close High Low Wei Premium (%) BTC % USD %
ANXBTC 315 314 321 296 314 4.84 856 0.53 255,944 0.52
BitFinex 315 314 320 304 312 4.26 17,511 11.25 5,580,123 11.25
BitKonan 315 320 328 301 310 6.12 3 0.00 899 0.00
BitStamp 315 313 320 306 314 4.33 8,803 5.87 2,887,517 5.87
BTC-e 312 311 315 304 311 3.79 5,990 4.09 1,883,258 4.05
HitBTC 320 321 331 311 320 4.29 370 0.25 123,580 0.25
itBit 315 313 317 305 313 2.94 1,485 1.11 528,609 1.11
Kraken 315 314 320 305 316 3.75 10 0.00 3,457 0.00
Lake 322 321 330 315 319 3.21 3,865 2.69 1,309,233 2.69
LocalBitcoins 386 381 1001 249 345 136.75 1,458 0.91 567,164 1.00
TheRock 317 313 335 312 316 2.05 10 0.00 3,051 0.00

BPI 315 314 319 308 3.55

BkCn 160,353 100.00 48,689,795 100.00
Abbreviations: BkCn, Blockchain; BPI, Bitcoin Price Index; H/L, high/low ratio; Wei, transaction-weighted price.

BkCn). The share of global bitcoin trades and the share of USD trades in global bitcoin markets are similar,

indicating that these exchanges all primarily trade in USD-BTC. The selected exchanges account for 26%

of the USD-BTC trade volume during this time period. According to Bitcoincharts12 , there are 72 active

exchanges globallythough many do not trade in USDthus, these eleven exchanges represent 15% of

all exchanges. BitFinex is the largest exchange, representing a daily median of 11.25% of global bitcoin

transactions, whereas the transaction volume of the smallest exchanges (BitKonan, Kraken, and The Rock

Trading) are negligible on the scale of the global bitcoin market.13

4 Investigating Law of One Price Between Exchanges

Each bitcoin price-series contains a time trend, so we use a cointegration test to verify whether LOOP is

satisfied. Given the price of bitcoin in two different markets, m1 and m2, a cointegration test determines if

any linear combination of the price-series results in a stationary residual, t :14

pm1
t pm2
t = t (1)

The cointegration test establishes whether a exists. If the two series do not cointegrate, then no linear

combination can be found that results in a stationary residual and therefore the two series cannot obey the

LOOP. If the two series cointegrate, LOOP requires that the long run linear cointegrating parameter, ,
12 bitcoincharts.com/markets/list
13 Like any currency, a bitcoin is divisible A trade volume of 3 bitcoin does not correspond to a maximum of three trades of

1 bitcoin each. Currently, the smallest unit is called 1 bitcoin satoshi, a hundred-millionth of a bitcoin (109 ). Each exchange
can determine the smallest trade it will accept.
14 This definition is adopted from the purchasing power parity example in Greene (2007), Chapter 20.

7
equals one: the two prices should be the same up to a constant.15 If it does not equal one, the two price-series

display a statistically significant drift over the sample period.

4.1 Method

We first verify that both price-series are unit root, employing two tests with opposing null hypotheses for

increased robustness. Our test with a null of the unit root is the Elliot et al. (1996) method of evaluating

the Dickey-Fuller test (Dickey and Fuller, 1979) using generalized least squares (DFGLS). The test with a

null of level stationarity is the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test (Kwiatkowski et al., 1992).

If both series are unit root, we apply the Johansen (1995) trace test to each at lag `. The Johansen

trace test iteratively examines the hypothesis that r or fewer cointegrating equations exist, with r = 0

representing no cointegration. With two series, a cointegrating relationship is the simultaneous result that

r 1 is significant and r = 0 is not. If the Johansen trace test finds cointegration, the cointegrating

parameter can be estimated using a vector error correction model (VECM) with the vector of the bitcoin

price pm,t for a given exchange m and the reference series rt , Pm,t = [pm,t , rt ]:

`1
X
Pt = 0 Pt1 + i Pti + + t (2)
i=1

where Pt = Pt Pt1 , and the market specific subscript on each term is omitted. The lag is first

selected according to the Schwarz Bayesian information criterion (SBIC). The residual of the VECM model

is evaluated for autocorrelation using a Lagrange Multiplier (LM) test as described in Johansen (1995). If

autocorrelation is found, the lag is increased by one and the tests are repeated.

If at least one series is not unit root at the selected lag, the Johansen trace test cannot be applied.

Instead, we apply the Pesaran et al. (2001) test (PSS). The PSS test allows for one series to be non-unit

root, but requires a single direction of granger-causality between the series. Formally, the PSS bounds test

is based on the unrestricted conditional error-correction model (CECM), expressed in notation of equation

2 as:

`
X
pyt = + i Pti + 0 log Pt1 + 0 pxt + t (3)
i=1

where pxt refers to the exchange that granger-causes the prices on exchange y, pyt .16 In cases where granger-

causality is confirmed, BitFinex is usually identified as the granger-causing series. Given the presence of

series that are not stationary, we use the methodology of Toda and Yamamoto (1995) (TY) to determine
15 It is expected that  6= 0 as each exchange has various fees, which should introduce an arbitrage band.
16 Or, y
in the words of Pesaran et al. (2001), px
t is the forcing function for pt .

8
Granger causality, first estimating a vector autoregression model (VAR) on the data with lags ` + 1:

`+1
X
Pt = c + i Pti + t (4)
i=1

where Pt(`+1) is an exogenous variable. If we find residual autocorrelation at either `, the lag is incremented

and the procedure repeated.

The long-run price-series relationship is captured in the coefficient = [1, ]. If = 1, the market and

the reference series perfectly share the same long-run trend. The typical size of the 99% confidence interval

around is 0.03, therefore [0.97, 1.03] will be interpreted as showing LOOP.

4.2 Long-Run Relationships

The results from the unit root test, cointegration test, and long run cointegrating relationship for each price-

series are displayed in Table 3 and 4.17 The rows labeled Unit Root?, Cointegrated?, and Law of One

Price? report whether the tests in the preceding rows confirm series unit root, series cointegration with

BitFinex, and Law of One Price with BitFinex respectively. We use BitFinex, the largest bitcoin exchange,

as the reference series. This does not imply that other exchanges are less valid or less important: BitStamp,

the second-largest exchange, has a longer time series than BitFinex; BTC-e, the third largest exchange, was

found to be a price leader by Brandvold et al. (2015). The CoinDesk Bitcoin Price Index (BPI), created

by Quandl, reports bitcoin prices averaged across selected exchanges and may therefore be more robust to

exchange-specific effects.18 We verify robustness using other series in section 4.3.1. The BitKonan Open,

LocalBitcoins Weighted, and The Rock Trading Low price suffer from residual autocorrelation for all lags

up to 15, indicated by ` > 15, and are therefore not tested. The BPI does not have a Transaction-weighted

price-series.

For any given exchange, LOOP findings in Tables 3 and 4 are the same for all tested price-series. The

cause of LOOP failure is therefore not a characteristic that can be isolated to a single price-series, but is a

general feature of an exchange. The price-series from HitBTC and LocalBitcoins consistently fail to attain

LOOP with BitFinex. HitBTC fails as it does not have a cointegrating relationship with BitFinex while

LocalBitcoins does cointegrate but does not obey LOOP.

Figure 2 is the difference between BitFinex and the series prices for LocalBitcoin and HitBTC the two
17 VECM postestimation test results for eigenvalue stability are not included, but are consistent with well behaved series:

There is one unit modulus, with all others strictly less than one.
18 There are several shortcomings of the BPI, variants of which are shared by most bitcoin price indexes: it only posts bitcoin

prices in US dollars, Euros, and Chinese yuan even though bitcoin is purchased using many currencies; the criteria for an
exchanges initial selection is unknown, as are the details of the price averaging procedure; the BPI only posts open, close, high,
and low prices, not a transaction-weighted price; and it does not provide trade volumes tailored to the price index it reports.
Once selected, an exchange in the BPI must satisfy criteria described at CoinDesk (www.coindesk.com/bitcoin-price-index/).

9
 Table 3: Price Cointegration with BitFinex: Open and Close Prices

ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock BPI
OPEN
Lag ` 4 > 15 2 1 3 1 1 1 1 6 1
Unit Root Test
DFGLS 0.30 0.34 0.31 0.18 0.29 0.27 0.29 -2.61 0.33 0.21
KPSS 7.17 11.9 17.7 8.51 17.8 17.9 17.7 16.2 5.2 17.8
Unit root? Yes Yes Yes Yes Yes Yes Yes No Yes Yes
Cointegration Test
r=0 89 81 110 14 136 202 111 35 110
r1 5.21 3.64 3.28 5.14 3.77 3.45 3.22 5.37 3.44
PSS 180
Cointegrated? Yes Yes Yes No Yes Yes Yes Yes Yes Yes
Long-Run Cointegrating Trend
1.00 1.00 0.99 1.00 1.01 1.00 1.14 0.98 1.00
LOOP? Yes Yes Yes No Yes Yes Yes No Yes Yes
CLOSE
Lag ` 1 3 7 5 3 2 4 5 4 1 5
Unit Root Test
DFGLS 0.35 -0.04 0.41 0.43 0.21 0.41 0.32 0.57 -0.49 0.29 0.41
KPSS 17.7 8.93 4.52 5.96 8.46 11.8 7.19 5.96 7.11 17.9 5.97
Unit root? Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Cointegration Test
r=0 263 83 30 30 15 84 53 32 90.37 208 29
r1 3.47 5.62 5.45 4.56 5.54 5.05 4.27 4.90 4.71 3.45 5.45
Cointegrated? Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes
Long-Run Cointegrating Trend
1.01 0.98 1.00 0.99 1.00 1.00 1.00 1.13 0.99 1.00
LOOP? Yes Yes Yes Yes No Yes Yes Yes No Yes Yes
Abbreviations: ANX, ANXBTC; BitK, BitKonan; BitS, BitStamp; DFGLS, Dickey-Fuller with Generalized Least Squares; KPSS,
Kwiatkowski-Phillips-Schmidt-Shin; Lake, LakeBTC; Local, LocalBitcoin; LOOP, law of one price; Rock, The Rock Trading.
Significance: 1%, **5%, *10%.

19
series that failed to yield LOOP. The HitBTC differences are not stationary, visually confirming the lack

of cointegration reported in Tables 3 and 4. Recall that LocalBitcoins cointegrates with BitFinex, but does

not have LOOP, consistent with the observable trends in Figures 2b, 2d, 2f, and 2g.20 The high volatility of

LocalBitcoinsdetected in price summaries presented Table 2is clearly visible.

4.3 Robustness

4.3.1 Other Exchanges As The Reference Series

We re-examine LocalBitcoins and HitBTC using the other exchanges as reference series, to evaluate whether

the failure of LOOP could result from the decision to use BitFinex as a reference series. We apply all the

methods of section 4.1, but unlike the detailed results in Tables 3 and 4, we report only the cointegrating

parameter in Table 5. An N C value indicates the Johansen test found no cointegration between the two

series, N GC indicates that a single direction of granger-causality could not be established when a series failed

the unit root test. Results in Table 5 confirm that failure to find LOOP does not stem from using Bitfinex
19 A trend line constructed using a Lowess smoother with a bandwidth of 40% is included to facilitate analysis of long run

trends.
20 For example, according the results in Table 4, the LocalBitcoins Low price difference, pLocal,Low pBitF inex,Low graphed
t t
in Figure 2g is not stationary (as required by LOOP), but pLocal,Low
t 0.89pBitF
t
inex,Low
would be.

10
 Table 4: Price Cointegration with BitFinex: Weighted, High, and Low Prices

ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock BPI
WEIGHTED
Lag ` 5 2 5 6 6 5 5 5 > 15 6
Unit Root Test
DFGLS 0.46 0.41 0.47 0.45 0.43 0.54 0.39 0.58 0.42
KPSS 5.96 11.9 5.99 6.13 4.91 6 6.02 5.97 5.21
Unit root? Yes Yes Yes Yes Yes Yes Yes Yes Yes
Cointegration Test
r=0 53.12 119.11 48.27 23.79 11.54 72.48 57.62 40.53 38.20
r1 4.62 3.95 5.34 4.08 4.82 5.57 5.60 4.91 4.27
Cointegrated? Yes Yes Yes Yes No Yes Yes Yes Yes
Long-Run Cointegrating Trend
1.00 0.99 1.01 0.99 1.00 1.00 1.00 0.98
LOOP? Yes Yes Yes Yes No Yes Yes Yes Yes
LOW
Lag ` 5 1 1 5 5 4 5 2 1 > 15 1
Unit Root Test
DFGLS 0.33 -0.22 0.10 0.27 0.21 0.29 0.25 0.29 -0.58 -1.23
KPSS 5.83 17.4 17.4 5.88 5.65 7.08 5.92 11.7 3.97 4.04
Unit root? Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Cointegration Test
r=0 44 150 185 39 16 66 53 103 229 251
r1 4.78 3.49 3.67 4.53 4.56 4.28 5.27 3.49 3.54 3.64
Cointegrated? Yes Yes Yes Yes No Yes Yes Yes Yes Yes
Long-Run Cointegrating Trend
0.98 0.98 1.00 0.99 1.00 1.00 1.00 0.89 1.00
LOOP? Yes Yes Yes Yes No Yes Yes Yes No Yes
HIGH
Lag ` 5 1 4 1 3 1 2 1 1 3 5
Unit Root Test
DFGLS 0.55 0.36 0.60 0.53 0.38 0.48 0.48 0.51 -8.19 0.34 0.58
KPSS 6.06 18 7.28 18 8.53 17.9 12.1 17.9 2.08 9.3 6.06
Unit root? Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes
Cointegration Test
r=0 76 248 48 129 14.01 133.45 115.57 107.99 60.69 36.84

r1 4.73 4.97 5.02 4.98 5.67 5.53 4.88 4.74 5.53 5.21
PSS 60.78
Cointegrated? Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes
Long-Run Cointegrating Trend
1.02 0.99 1.00 0.98 0.99 1.00 1.00 1.06 0.97 0.99
LOOP? Yes Yes Yes Yes No Yes Yes Yes No Yes Yes
Abbreviations: ANX, ANXBTC; BitK, BitKonan; BitS, BitStamp; DFGLS, Dickey-Fuller with Generalized Least Squares;
KPSS, Kwiatkowski-Phillips-Schmidt-Shin; Lake, LakeBTC; Local, LocalBitcoin; LOOP, law of one price; Rock, The Rock Trading.
Significance: 1%, **5%, *10%.

as the reference series. In all 90 cases where the reference series had LOOP with BitFinex, LocalBitcoins

and HitBTC do not yield LOOP.21

We can establish market segmentation if the markets that do not satisfy LOOP with the reference series

obey LOOP with each other. The relationship between LocalBitcoins and HitBTC exchanges is subtle: They

are always cointegrated with each other, though they never obey LOOP with each other. Therefore, even

though their price trends have more in common with each other than with BitFinex, they do not represent

their own segmented market.

21 LOOP results are transitive: all exchanges that adhere to LOOP with BitFinex also enjoy LOOP with each other.

11
 Figure 2: Price difference for series that failed LOOP

(a) Open, HitBTC (b) Open, Local (c) Close, HitBTC (d) Close, Local

(e) Low, HitBTC (f) High, Local (g) Low, Local (h) High, HitBTC (i) Weighted, HitBTC

Table 5: Consistency of LOOP failure using other reference series

Series with BitFinex LOOP Series w/o BitFinex LOOP

BitFinex ANX BitK BitS BTC-e itBit Kraken Lake Rock BPI HitBTC Local
LocalBitcoin
Open 1.14 1.14 NGC 1.14 1.16 1.14 1.13 1.14 1.15 1.15 1.21
Close 1.13 1.13 1.13 1.13 1.14 1.13 1.14 1.13 1.15 1.14 1.20
Low 0.89 0.91 0.91 0.90 0.91 0.90 0.89 0.89 0.90 0.90 0.95
High 1.06 NGC 1.03 1.07 NGC NGC NGC NGC NGC NGC NGC

HitBTC
Open NC NC ` > 15 N C NC NC NC NC NC NC 1.21
Close NC NC NC NC NC NC NC NC NC NC 0.84
Weighted NC NC NC NC NC NC NC NC NC ` > 15
Low NC 0.95 NC NC 0.94 NC 0.92 NC ` > 15 NC 1.06
High NC NC 0.94 NC NC 0.94 NC NC NC NC NGC
All observations are significant at 1%
Abbreviations: ` > 15, Lag to avoid residual autocorrelation greater than 15 days; NC, Not Cointegrated; NGC, Could not
establish single direction granger-causality.

4.3.2 Structural Breaks

Visual examination of the series presented in Figures 1 and 2 reveals potential structural breaks occurring

between mid-December, 2014 and mid-January, 2015. There is no known major event occurring within this

time-period in bitcoin history. To verify that cointegration failure is not a result of a structural break,

the sample is split into two sub-periodsthe first 200 observations preceding 17 December 2014 (the first

segment), and the final 176 observations after 5 February 2015 (the second segment). We again apply

the methods from section 4.1 to each sub-segment. This division separates the decreasing and increasing

sections of the price-series for all exchanges, as well as the observed hump for HitBTC in Figure 2, while

retaining same-size sample within each segment across all exchanges. Results for all series are presented in

Table 6, using BitFinex as the reference series and the same notation as Table 5.

The results for HitBTC show that failure of LOOP in the full sample is caused by the behavior of the

12
 Table 6: Structural Break Cointegration with BitFinex

ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock BPI
Open
Full Sample 1.00 ` > 15 1.00 0.99 NC 1.00 1.01 1.00 1.14 0.98 1.00
First Segment 1.00 0.96 0.99 0.99 0.99 1.00 0.98 1.00 1.15 0.98 0.99
Second Segment 1.01 1.01 0.99 0.91 0.84 0.99 1.01 0.97 0.88 0.93 0.98

Close
Full Sample 1.01 0.98 1.00 0.99 NC 1.00 1.00 1.00 1.12 0.99 1.00
First Segment 1.00 0.95 0.99 0.99 0.99 1.00 0.98 0.99 NGC 0.99 0.99
Second Segment 1.01 1.04 0.99 0.91 NC 1.00 0.99 0.97 1.35 0.92 0.98

Weighted
Full Sample 1.00 0.99 1.00 0.99 NC 1.00 1.00 1.00 ` > 15 0.98
First Segment 1.01 0.97 1.00 0.99 1.00 1.00 0.99 1.00 1.05 0.98
Second Segment NC 1.04 1.00 0.91 NC 0.99 1.01 0.97 1.04 0.93

Low
Full Sample 0.98 0.98 1.00 0.99 NC 1.00 1.00 1.00 0.89 ` > 15 1.00
First Segment 0.99 0.96 0.99 0.99 0.99 1.01 0.98 1.00 0.88 1.00 0.99
Second Segment 1.02 1.01 1.01 0.92 NC 0.98 0.98 0.96 0.80 0.86 NC

High
Full Sample 1.02 0.99 1.00 0.98 NC 0.99 1.00 1.00 NGC 0.97 0.99
First Segment 1.03 0.97 0.99 0.99 1.00 1.00 0.98 0.99 NGC 0.95 0.99
Second Segment 1.02 1.02 0.99 0.89 NC 1.00 1.00 0.97 NC 0.97 0.97
All observations are significant at 1%
Abbreviations: ` > 15, Lag to avoid residual autocorrelation greater than 15 days; NC, Not Cointegrated;
NGC, Could not establish single direction granger-causality.

second segment; the first segment does satisfy LOOP. There are no publicly stated reasons that could explain

this turning point for HitBTC. This is additional evidence that failure of LOOP is not caused by market

segmentation between exchanges: LocalBitcoins fails LOOP in both segments.22

BitKonan and Rock Trading illustrate that structural issues such as autocorrelation (indicated by ` > 15

in the full sample) may point to trend changes within the sample (the Low price result for Rock Trading shows

a clear change in the cointegration parameter for the two periods), but not necessarily so (the parameter

for the LocalBitcoin Transaction-weighted price is relatively stable). This could reflect limitations inherent

in the significantly reduced sample size, diminishing our ability to establish additional conclusions.

The differences in the parameter for the shorter sample is greater than the full series, making verification

of LOOP harder. The second segment of BitKonan Low prices clearly fails LOOP with a of 0.85, but how

should we regard LakeBTCs Low of 0.96? Given that the 99% confidence interval around LakeBTC is

[0.93,0.99], a = 1 is statistically unlikely. We shall classify an exchange containing at least one segment

with
/ [0.90, 1.10] as displaying Strong LOOP Failure, while an exchange containing [0.90, 1.10]

but
/ [0.97, 1.03] displays Weak LOOP Failure. There are therefore four exchanges that have no

LOOP failure: ANXBTC, BitStamp, itBit, and Kraken, as well as the BPI. Two exchanges (LakeBTC and
22 We emphasize that data from LocalBitcoins and HitBTC should not be avoided per se, rather that the nature of bitcoin

exchanges requires initial testing to verify that a price-series is representative.

13
BitKonan) have Weak LOOP Failure, while two more (The Rock Trading and BTC-e) have Strong LOOP

Failure. LocalBitcoins and HitBTC display Consistent LOOP Failure as they do not achieve LOOP over

the full sample.

5 Exchange Characteristics and Bitcoin Pricing Behavior

The preceding section identifies the extent to which bitcoin exchanges may or may not be representative

of bitcoin price trends. In this section, we examine the extent to which differences in prices may be explained

by characteristics of the exchanges.

5.1 Exchange Characteristics

Table 7 summarizes select characteristics of the 11 exchanges across three categories: fees, exchange trans-

parency, and regulations. Additional notes are in italics. BPI is not an exchange and is not included.

Table 7: Characteristics of Bitcoin Exchanges

Fees Exchange Transparency Regulations
Name Trading Deposit Withdrawal Incorp. Physical Server AML/KYC ID required at

ANXBTC Free Free $30 Hong Kong Hong Kong Hong Kong Yes Funding
Offers a debit card. A fully centralized service: upon indicating a quantity to buy/sell, users receive a 30-second take-it-or-leave-it price
offer. Also accepts EUR, GBP, CHF, CAD, AUD, YEN, HKD, NZD, and SGD.

BitFinex 0.00%-0.20% $20 $20 British Virgin Islands UK Italy Yes W-USD
Advertising a job opening at London office. Claims offices in New York,Hong Kong and Taiwan. Included in BPI. Free Bitcoin deposit

BitKonan 0.29% $15 $15 Croatia Croatia Netherlands Yes None

Accepts deposits in multiple currencies, but automatically converts into USD.

BitStamp 0.10%-0.25% $7.50 $15 UK UK, USA, Luxembourg (?) Yes Funding
Offers a debit card. Included in the CoinDesk BPI. Hacked during sample period. Also accepts EUR, GBP.

BTC-e 0.20% $0-$60 $0-$60 Cyprus (?) (?) Yes None

Included in the CoinDesk BPI. Known for an active message board displayed alongside the buy/sell window. Fees may vary by user.
Also accepts EUR, RUR.

HitBTC 0.10% $9 $9 (?) (?) (?) Yes (Tiered) W-USD

No countries are referenced in the Terms of Service Agreement. Also accepts EUR.

itBit 0.10%-0.50% $10 $20 USA USA USA Yes Funding

Board of directors includes former FDIC chairman, Sheila Bair. Also accepts EUR, SGD.

Kraken 0.10%-0.35% $20 $20 USA USA (?) Yes Funding

Also accepts EUR, GBP, JPY. Does not allow off-chain transactions.

LakeBTC 0.00%-0.20% Free $6-$8 (?) China (?) Yes None

Terms of Service Agreement refers to USA, Sweden and the UK. Also accepts CNY, SEK.

LocalBitcoins 0.00%-1.00% N/A N/A Finland Finland (?) No None

Acts as a message board and bitcoin wallet to facilitate peer-to-peer trades. Allows prices to be posted in any currency or
commodity (gold,etc.), but only accepts bitcoin deposits into wallets. Has a fee to deposit bitcoin. Database hacked during sample period.

The Rock Trading 0.075%-0.50% Free e11 Malta Malta (?) Yes (Tiered) W-USD
Also accepts EUR, GBP
Abbreviations: AML, anti-money laundering; Incorp., country of incorporation; KYC, know-your-client/customer; Tiered: Has tiered KYC compliance

: The fee for a transfer of US$1,000 via bank wire either into (deposit) or out of (withdrawal) the exchange.

The Fees columns summarize various fees across the exchanges. Fees typically depend on trade volume,

with most trading fees between 0% to 0.50%. LocalBitcoins is unusual in that it charges no fee for buying,

14
and a 1% fee for selling bitcoin, regardless of volume. Deposit and withdrawal fees are calculated for a

transfer of US$1,000 via bank wire.23 LocalBitcoinswhich does not accept currency depositshas no

currency transfer fees. Most exchanges charge a higher fee for withdrawals than deposits.

The next section in Table 7 represent measures of an exchanges transparency: is it even possible to

determine which countrys laws the exchange should follow? The country of incorporation (Incorp.) is

determined by the laws referenced in the exchanges Terms of Service. The physical location (Physical)

reflects either a mailing address or the location of an advertised job opening. Job location may be misleading

as employees may telecommute, while the mailing address may be for a holding company. Finally, we also

report server location (Server) as this may be a better indicator of the physical location of the exchanges

headquarters. It is rare that a country of incorporation cannot be determined. The physical address usually

reflects the country of incorporation, although it may be different (BitFinex). Most exchanges refuse to

reveal their server locations. The available data reveals the global diversity of bitcoin exchanges, with 11

exchanges revealing known connections to 12 countries.

The final pair of columns (Regulations) are measures of an exchanges adherence to AML/KYC. The

functional details required to adhere to AML/KYC may vary, but broadly speaking KYC adherence involves

collecting data to verify the identity of an account holder (drivers license, passport, etc.), while AML requires

that an exchange monitor transaction data to identify suspicious behavior. AML/KYC indicates whether

an exchange has a statement that implies that it adheres to AML/KYC. Some exchanges will state compliance

to only AML policies, but their description of enforcement includes collecting passport or ID photos, which

we interpret as evidence of a KYC policy, even if it not stated as such.

It is possible that an exchange claims it adheres to an AML/KYC policy, without effectively implementing

one. While we cannot verify AML compliance, we can test an exchanges KYC adherence. We open accounts

at each exchange in the sample to identify when (and if) an exchange attempts to gather KYC information.

We make no attempt to hide the geographic location of the computer (USA) and indicate USA residency if

prompted. Beyond the email address used to establish the account, we provide no additional information to

the exchange. We attempt to fund each account with 0.01 bitcoin (approximately USD $6), sell the bitcoin

for USD, buy a replacement bitcoin on the exchange, and withdraw both USD and bitcoin.24

Four exchangesANXBTC, BitStamp, itBit, and Kraken do not allow any transfers or transactions

without identity verification, indicated by Funding in the final column in Table 7. The remaining ex-
23 Alternatives to bank wires that are accepted at somebut not allexchanges are virtual payments (i.e., PayPal) or credit

card funding (VISA, MasterCard). The bank wire fee may change based upon the value transferred.
24 All bitcoin are sent from the same wallet at the exchange Coinbase, which is not an exchange in the study. We initiate

(but do not complete) the procedure to withdraw the USD generated by the bitcoin sale due to legal and security concerns.
USD withdrawals cannot occur on LocalBitcoins as it does not provide a platform for the transfer of currency. We do not sell
the transferred bitcoin on LocalBitcoin due to legal and security concerns.

15
changes allow the account to be funded with the bitcoin, sold for USD, and a replacement bitcoin to be

purchased and withdrawn from the exchange. These exchanges differ, however, in regards to withdrawing

USD. Three exchangesBitKonan, BTC-e, and LakeBTC would allow us to withdraw USD without pro-

viding additional information. It is possible that these three exchanges have a limit below which they do

not require ID, and that the amount of USD we were attempting to withdraw was below this limit.

Exchanges differ in other, less quantifiable ways, noted in Table 7. The currencies accepted for bitcoin

purchases vary across exchanges. ANXBTC and BitStamp offer a debit card. In addition to AML and KYC

compliance, itBit secured a charter from the New York State Department of Financial Services and have

appointed former FDIC chair Sheila Bair to their board of directors. Only one exchange, Kraken, states

that it does not allow off-chainingallowing transactions to complete prior to their integration into the

blockchain. The remaining exchanges do not address when or how they reconcile off-chain transactions to

the blockchain.

Most exchanges do not hold their own inventory of either currency or bitcoin, instead they match user

with a buy (sell) to a user with an offer (bid). ANXBTC users, on the other hand, submit the number of

bitcoin they wish to trade, and then have 30 seconds to accept a price offered by the ANXBTC exchange.

In contrast, LocalBitcoins allows users to transfer bitcoin to another users wallet, but does not provide a

platform for the reciprocal transfer of currency.25 Instead, a LocalBitcoins user posts an offer with price

and payment specifications on a LocalBitcoins posting board, which aggregates and lists all individual offers

by username, price, and zip code. Examples of payment specifications include making a cash deposit at a

specific bank branch, PayPal transfer, or a face-to-face meeting for a commodity swap (such as gold). Stop

and limit orders are available on all exchanges except LocalBitcoin. Because it uses a posting board system,

LocalBitcoins is one of the few exchanges where one user can single out another user with whom to conduct

their transaction.

5.2 Price trend and Exchange Characteristics

In Section 1 we argued that it is not unreasonable to expect bitcoin to adhere to LOOP, and that LOOP

failure of an exchange must result from a characteristic of that exchange rather than an inherent property

of bitcoin. To evaluate which characteristics may contribute to our finding of non-LOOP exchanges, Table

8 groups exchanges by LOOP results, and summarize as indicator variables the exchange characteristics

contained in Table 7 and trade volumes from Table 2. For example, if the exchange posts an AML/KYC

statement Stated AML/KYC takes a value of 1. If an exchange requires additional information for a user

to withdraw currency (whether that in the funding stage or the withdrawal stage), the ID for W-Currency
25 As the LocalBitcoin platform does not transfer fiat currency, it maintains that it does not need to comply with AML/KYC.

16
takes a value of 1.

Free Fiat Deposits an indicator variable for fees, as it is uniform across all depositors regardless of

currency, account size, transaction size, or transaction frequency. LocalBitcoins is the only exchange that

does not have free bitcoin deposits. We also indicate whether exchanges have fixed trading fees (Fixed

Trading Fee): A sliding trading fee typically rewards larger trades which may distort the market, especially

if large trades allows users to avoid paying a trade fee altogether (Free Trades Feasible). Inclusion in

the BPI is a proxy for an exchanges reputation in the bitcoin community. Several exchanges allow trades

in currencies other than USD. Only Kraken states that it does not allow off-chain transactions, and only

LocalBitcoins allows users to select who they wish to trade with (Targeted Trading). If any of the three

locations associated with the exchange in located within the USA (USA location?), the exchange is legally

required to comply with AML/KYC.

Table 8: Cointegration and Exchange Properties

Exchanges with LOOP Weak LOOP Fail Strong LOOP Fail Consistent LOOP Fail
ANX BitS itBit Kraken BitK Lake BTC-e Rock HitBTC Local

Exchange Transparency (Table 7)

Known incorporation? 1 1 1 1 1 0 1 1 0 1
Known physical address? 1 1 1 1 1 1 0 1 0 1
Known server location? 1 1 1 0 1 0 0 0 0 0
USA location? 0 1 1 1 0 0 0 0 0 0

Regulation Compliance (Table 7)

Stated AML/KYC? 1 1 1 1 1 1 1 1 1 0
ID for Funding 1 1 1 1 0 0 0 0 0 0
ID for W-BTC 1 1 1 1 0 0 0 0 0 0
ID for W-Currency 1 1 1 1 0 0 0 1 1

Fees (Table 7)
Free Fiat Deposit 1 0 0 0 0 1 0 1 0
Free Bitcoin Deposit 1 1 1 1 1 1 1 1 1 0
Fixed Trading Fee 1 0 0 0 1 0 1 0 1 0
Free Trades Feasible 1 0 0 0 0 1 0 0 0 1

Share of Global Transactions (Table 2)

High (> 4%) 0 1 0 0 0 0 1 0 0 0
Low ( 0%) 0 0 0 1 1 0 0 1 0 0

Other Characteristics (Table 2)

Included in BPI 0 1 0 0 0 0 1 0 0 0
EUR trade 1 1 1 1 0 0 1 1 1 1
GBP trade 1 1 0 1 0 0 0 1 0 1
Non EUR/GBP fiat 1 0 1 1 0 1 1 0 0 1
Hacked 0 1 0 0 0 0 0 0 0 1
Debit Card Available 1 1 0 0 0 0 0 0 0 0
Does not allow off-chaining 0 0 0 1 0 0 0 0 0 0
Targeted Trading 0 0 0 0 0 0 0 0 0 1
Abbreviations: AML, anti-money laundering; ANX, ANXBTC; BitK, BitKonan; BitS, BitStamp;KYC, know-your-customer;
Local, LocalBitcoin; LOOP, relative law of one price; Rock, The Rock Trading.

The two exchanges that Consistently Fail LOOP share seven characteristics: neither disclose their server

location; they are not located in the USA; they allow funding of accounts and withdrawal of BTC without

17
ID; neither have an extreme trade share (either high or approximately zero share of global transactions);

neither have a statement banning off-chain transactions; and neither have a debit card available. Of these

seven characteristics, six are found in at least one of the LOOP exchanges server location (Kraken), USA

location (ANXBTC), no extreme high/low volume (ANXBTC, Kraken), off-chaining (all), and debit card

(itBit, Kraken) so it seems unlikely that these characteristics can explain why LOOP fails.26 This leaves

only the KYC implementation variables: all exchanges with LOOP require IDs to Fund the account, a

requirement that neither HitBTC nor LocalBitcoin enforce.

The exchanges that exhibit Inconsistent LOOP Failure (Weak or Strong) also do not require identification

information to fund the account, so we interpret a failure to require an ID to fund as predictive of LOOP

failure. There is no characteristic that distinguishes exchanges with Consistent LOOP Failure from those

that have Inconsistent LOOP Failure. Inconsistent LOOP exchanges differ from each other on acceptance

of Euro trades, yet it seems unlikely that these are a distinguishing feature as both LOOP and Failed

LOOP exchanges allow Euro trades. This implies that there are additional, potentially hidden, exchange

characteristics not captured on Table 8 that explain the degree of LOOP failure.

5.3 Price Deviation and Exchange Characteristics

We define the price deviation of an exchange as the difference between a given price-series and the corre-

sponding price-series for the reference exchange (BitFinex):

pit
 
Deviationit
 
= log  BitF inex 1
 (5)
p t

For some series, like the High price-series, a deviating price is typically greater than the reference series
pit
 
BitF
pt inex > 1 so a positive coefficient would indicate more deviation, while for others, like the Low price
pi
 
series, deviating prices are typically less than the reference series pBitFtinex < 1 and a negative coefficient
t

indicates more deviation. By using the absolute value of the deviation from parity in equation 5, a positive

coefficient can consistently be interpreted as greater price deviation from the reference series.27 We also

calculate the daily Trade Volume (T V Rti ) for each exchange, normalized by the daily BitFinex volume to

account for a growth in bitcoin trading over the sample

V olti
T V Rti =
V oltBitF inex
26 It is important to recall that we only selected exchanges reporting 410 days or more with transactions; the finding regarding

trade volume can therefore more accurately be considered as suggesting that active exchanges follow long-run price trends
regardless of size.
27 The log is applied to normalize the distribution.

18
 Table 9: Mean and Standard Deviation of Price Deviation and Normalized Trade Volume

No LOOP Failure Weak LOOP Failure Strong LOOP Failure LOOP Failure
ANX BitS itBit Kraken Lake BitK BTC-e Rock Local HitBTC Aggregate
Open -5.47 -6.08 -5.70 -4.85 -5.41 -4.35 -4.59 -4.51 -2.02 -4.05 -4.70
(1.38) (1.26) (1.18) (1.36) (1.05) (1.06) (0.93) (1.12) (1.17) (1.64) (1.64)
Close -5.32 -6.07 -5.72 -4.77 -5.39 -4.09 -4.64 -4.61 -2.09 -3.99 -4.67
(1.31) (1.27) (1.20) (1.25) (1.00) (1.11) (1.00) (1.17) (1.14) (1.59) (1.62)
Weighted -5.08 -6.00 -5.44 -5.02 -5.39 -4.23 -4.60 -4.56 -2.24 -4.01 -4.65
(1.22) (1.20) (1.20) (1.36) (1.07) (1.17) (0.93) (1.07) (0.28) (1.59) (1.52)
High -4.99 -5.77 -5.23 -4.79 -5.32 -4.09 -4.50 -4.14 0.78 -4.10 -4.21
(1.31) (1.24) (1.04) (1.24) (1.08) (1.31) (0.87) (1.28) (0.65) (1.77) (2.13)
Low -4.88 -5.75 -5.52 -4.92 -5.17 -4.16 -4.64 -4.71 -2.06 -3.93 -4.57
(1.44) (1.27) (1.36) (1.26) (1.05) (1.19) (1.06) (1.34) (0.50) (1.52) (1.58)
TVR 0.08 0.63 0.16 0.00 0.30 0.00 0.41 0.00 0.10 0.04 0.17
(0.10) (0.38) (0.17) (0.00) (0.27) (0.00) (0.24) (0.00) (0.09) (0.06) (0.27)

The mean and standard deviation for each price-series and exchange are summarized in Table 9. Price

deviations for exchanges that achieve LOOP are below average, while exchanges with Consistent LOOP

Failure have more deviations than the average. Daily trade volume varies over the four categories, with the

exchanges that have a negligible share of global bitcoin trade (Kraken, BitKonan, and Rock) also have a

negligible share of trade relative to BitFinex.

The relationship between LOOP failure and the magnitude of price deviations is evaluated using an an

OLS regression with robust standard error. Table 10 reports the coefficients.28 Prices from the BPI are

removed as it is an index constructed using prices from exchanges in this sample, not an exchange. BitFinex

is removed as it is the reference series. Date fixed effects are included.

Table 10: Price Deviation and LOOP Results

LocalBitcoins Included LocalBitcoins Removed

Open Close Weighted High Low Open Close Weighted High Low

Weak LOOP Failure 0.65*** 0.74*** 0.58*** 0.49*** 0.61*** 0.65*** 0.74*** 0.58*** 0.49*** 0.61***
Strong LOOP failure 0.98*** 0.84*** 0.80*** 0.87*** 0.59*** 0.98*** 0.84*** 0.80*** 0.87*** 0.59***
Consistent LOOP Failure 2.49*** 2.43*** 2.26*** 3.53*** 2.27*** 1.48*** 1.48*** 1.37*** 1.09*** 1.33***

Constant -5.53*** -5.47*** -5.38*** -5.19*** -5.27*** -5.53*** -5.47*** -5.38*** -5.19*** -5.27***

Date fixed effect? Y Y Y Y Y Y Y Y Y Y

Adj. R2 0.30 0.30 0.31 0.36 0.30 0.14 0.14 0.14 0.10 0.14
N 4243 4243 4238 4241 4233 3817 3818 3812 3815 3807
Significance: 1%, **5%, *10%. [95% confidence interval] .

Two observations can be made from Table 10: The positive coefficients on all three dummy variables show

that LOOP failure is associated with larger price deviations. Secondly, the magnitude of price deviations and

the severity of LOOP failure are positively associated for all price-series except the Low prices. For the Low

price-series the magnitude of deviations associated with Weak LOOP Failure is slightly larger than Strong

LOOP Failure. Table 2 showed that LocalBitcoins prices are more volatile than other exchanges; excluding
28 The regression is Deviationit = 1 Weak LOOP Failure + 2 Stong LOOP failure + 3 Consistent LOOP failure + constant

19
it from the regression does not change the findings, though it does reduce the magnitude of deviations

associated with Consistent LOOP failure. Even with this reduction, the magnitude of price deviations on

exchanges that Consistently Fail LOOP are twice as large as the magnitude of deviations for exchanges that

have Weak LOOP Failure. Understanding which of the 22 exchange characteristics (summarized in Table

8) contribute toward price deviations can therefore help us identify characteristics associated with LOOP

failure.

Table 11 reports the results from an OLS regression of price deviation and exchange characteristics with

robust standard errors,29

X X
Deviationt = + k I k + V T V Rti + t F Et +  (6)
k t

where I k are the exchange characteristics indicated in Table 8, as well as their interactions, and F Et is a

date fixed effect.

Four of the 22 potential characteristics act as fixed effects for specific exchanges and are not considered: A

Stated AML/KYC policy, Free Bitcoin Deposit, and Targeted Trading all act as fixed effects for LocalBitcoin,

while Not Allowing Off-chaining is a fixed effect for Kraken. An additional six Known Incorporation,

Known Physical Address, High Volume Trade, Inclusion in the BPI, Hacked, and Debt Card Available

differ for only two exchanges, and frequently cause multicollinearity issues if included. Two more, Free Fiat

Deposits and ID For Currency Withdrawals, are problematic as they cannot be evaluated for LocalBitcoins

which does not allow fiat currency deposits. Finally, the values for ID for BTC withdrawals duplicates the

values of ID for Funding. This leaves nine of the original 22 characteristics for consideration.

Results in section 5.2 showed that requiring an ID to fund an account is an important characteristic in

explaining LOOP failure, and therefore ID to Fund must be included. Indicator variables for Knowledge

of Server Location, USA Location, Trades in Euros, GBP, and Trades in Other Currencies all result in a

variance-inflation factor above 3, indicating multicollinearity when included with ID to Fund, so they are

removed. This leaves 4 characteristics for consideration with price deviation: ID for Funding, Fixed Trading

Fee, Free Trades Feasible, and Low Global Share of bitcoin trade volume.

The basic regression, (i), for each price-series considers only the impact that requiring an ID to Fund has

on price deviations. Regression (ii) uses the remaining characteristics, trade volume, and their interactions

if they do not result in multicollinearity. All statistically significant variables have consistent signs across all

price-series, confirming results from Section 4.2 that the effects documented in this paper are not isolated to

a single price series.

29 Allowing for lags with Newey West Errors does not change results.

20
 Table 11: Price Deviation and Exchange Characteristics
Open Close Weighted High Low
(i) (ii) (i) (ii) (i) (ii) (i) (ii) (i) (ii)

ID to Fund -1.38*** -1.92*** -1.34*** -1.79*** -1.22*** -1.53*** -1.64*** -2.27*** -1.16*** -1.53***
ID to Fund x Low Share 1.50*** 1.50*** 0.98*** 1.25*** 1.18***

Low Global Share -0.40*** -0.23** -0.32*** -0.57*** -0.32***

Fixed Cost to Trade -0.12* 0.05 0.01 -0.83*** 0.09
Free Trade Feasible 0.83*** 0.82*** 0.78*** 1.86*** 0.99***
Trade Volume -0.33*** -0.39*** -0.58*** -0.74*** -0.33**

Volume x Low Share -167.55*** -248.08*** -159.61*** -160.24*** -227.45***

Volume x Fixed Cost -1.02*** -1.07*** -0.73*** -0.74*** -1.07***
Volume x Free Trade -3.06*** -2.56*** -2.47*** -5.07*** -2.72***

Constant -4.15*** -3.87*** -4.13*** -3.99*** -4.17*** -3.99*** -3.55*** -3.03*** -4.10*** -4.01***
Date fixed effect? Y Y Y Y Y Y Y Y Y Y
2
Adj. R 0.15 0.25 0.15 0.25 0.15 0.24 0.09 0.30 0.13 0.24
N 4243 4243 4243 4243 4238 4238 4241 4241 4233 4233
Significance: 1%, **5%, *10% .

Exchanges that require ID to Fund have smaller price deviations those than those that do not, though

the effect is diminished for exchanges with a low global share of bitcoin trades. Having a larger trade volume

is also associated with smaller deviations, an effect that is especially potent for the small exchanges.30

Exchanges where free trade is feasible usually as a result of trading large volumes see larger price

deviations, although this is mitigated in exchanges with larger trade volumes. It is worth noting this suggests

that fees on bitcoin exchanges are substantial enough to impact prices.

Of the characteristics considered above, requiring an ID to Fund an account have largest impact on

decreasing price deviations. This is consistent with findings that showed ID to be a good indicator of LOOP.

6 Conclusions

Crypto-currencies like Bitcoin are a comparatively new financial mechanism, and are not yet assimilated into

a consistent global regulatory framework. Of the 11 bitcoin exchanges we evaluate, we find that exchanges

which do not fully implement AML and/or KYCtwo regulatory policies already in broad use across the

global financial systemexhibit statistically different price patterns from exchanges that do. This suggests

two things. First, standard financial regulations can affect bitcoin markets, even across a group of different

global exchanges. Second, bitcoin-pricing data can differ across exchanges, not obeying the Law of One

Price, in ways that are not always obvious, demanding a deliberate approach to data sourcing.

It may seem strange that requiring an ID to fund an account should matter to bitcoin users. Although

bitcoin transactions are more private than cash transactions they are actually less anonymous. Every bitcoin

wallet is identified by a unique private key. The private keys of both wallets in a bitcoin transaction are
30 Recall that Low Global Share was defined as an exchange that traded approximately 0% of global bitcoin trades, so a

coefficient of 167.55 has a substantially smaller impact than it appears.

21
recorded on the blockchain. The blockchain can be thought of as a digital ledger, maintaining a history of

bitcoin transactions. Anonymity results only if it is impossible to prove a connection between the private key

and an individual.31 There is some evidence that users with criminal intentions use bitcoin: using Google

search data, Wilson and Yelowitz (2015) find evidence of significant criminal interest in bitcoin. For these

users, the anonymity promised by using bitcoin is its primary benefit, a benefit which is eliminated (or

substantially undermined) by exchanges which require ID.

References

Akram, Q. F., Rime, D., Sarno, L., December 2008. Arbitrage in the Foreign Exchange Market: Turning on

the Microscope. Journal of International Economics 76 (2), 237253.

Ali, R., Barrdear, J., Clews, R., Southgate, J., Q3 2014. The economics of digital currencies. Bank of England

Quarterly Bulletin.

Badev, A., Chen, M., 2014. Bitcoin: Technical background and data analysis. FRB Finance and Economics

Discussion Series 104.

Bohme, R., Christin, N., Edelman, B., Moore, T., Spring 2015. Bitcoin: Economics, technology, and gover-

nance. Journal of Economic Perspectives.

Brandvold, M., Molnar, P., Vagstad, K., Valstad, O. C. A., May 2015. Price discovery on bitcoin exchanges.

Journal of International Financial Markets, Institutions and Money 36, 1835.

Briere, M., Oosterlinck, K., Szafarz, A., 2015. Virtual currency, tangible return: Portfolio diversification

with bitcoins. Journal of Asset Management.

Chu, J., Nadarajah, S., Chan, S., July 2015. Statistical analysis of the exchange rate of bitcoin. PLOS ONE

10 (7).

Ciaian, P., Rajcaniova, M., dArtis Kancs, 2016. The economics of BitCoin price formation. Applied Eco-

nomics 48 (19), 17991815.

Dickey, D. A., Fuller, W. A., 1979. Distribution of the estimators for autoregressive time series with a unit

root. Journal of the American Statistical Association 74, 427431.

Dwyer, G. P., April 2015. The economics of Bitcoin and similar private digital economies. Journal of Financial

Stability, 8191.
31 For a discussion of the technology involved, refer to Badev and Chen (2014) or Bohme et al. (2015).

22
Elliot, G., Rothenberg, T., Stock, J., 1996. Efficient tests for an autoregressive unit root. Econometrica 64,

813836.

Eun, C. S., Sabherwal, S., April 2003. Cross-Border Listings and Price Discovery: Evidence from U.S.-Listed

Canadian Stocks. The Journal of Finance 58 (2).

Glaser, F., Zimmermann, K., Haferkorn, M., Weber, M. C., Siering, M., 9-14 June 2014. Bitcoinasset

or currency? Revealing users hidden intentions. Twenty-Second European Conference on Information

Systems (ECIS), Tel Aviv.

Greene, C., Shy, O., September 2014. E-cash and virtual currency as alternative payment methods. Journal

of Payments Strategy and Systems 8 (3).

Greene, W. H., 2007. Econometric Analysis. Pearson Eduction.

Halaburda, H., Gandal, N., October 2014. Competition in the cryptocurrency market. NET Institute Working

Paper No. 14-17Available at SSRN, http://dx.doi.org/10.2139/ssrn.2506463.

Johansen, S., 1995. Likelihood-Based Inference in Cointegrated Vector Autoregressive Models. Oxford Uni-

versity Press.

Kaplanov, N. M., 2012. Nerdy Money: Bitcoin, the Private Digital Currency, and the Case Against its

Regulation. Loyola Consumer Law Review 25 (1).

King, D., February 2016. Banking Bitcoin-Related Businesses: A Primer for Managing BSA/AML Risks.

Retail Payments Risk Forum Working Paper - Federal Reserve Bank of Atlanta.

Kristoufek, L., 2013. Bitcoin meets Google Trends and Wikipedia: Quantifying the relationship between

phenomena of the Internet era. Scientific Reports 3, article no. 3415.

Kroll, J., Davey, I. C., Felten, E. W., June 2013. The economics of bitcoin mining, or bitcoin in the presence of

adversaries. The Twelfth Workshop on the Economics of Information Security (WEIS 2013), Washington.

Kwiatkowski, P. D., Phillips, P., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationarity

against the alternative of a unit root: How sure are we that economic time series have a unit root?

Journal of Econometrics 54, 159178.

Lo, S., Wang, J. C., September 2014. Bitcoin as money? Federal Reserve Bank of Boston: Current Policy

Perspectives 14-4.

23
Luther, W., 2016. Cryptocurrencies, Network Effects, and Switching Costs. Contemporary Economic Policy

34 (3).

Nakamoto, S., 2008. Bitcoin: A peer-to-peer electronic cash systemAvailable at

http://bitcoin.org/bitcoin.pdf.

Parkinson, M., 1980. The extreme value method of estimating the variance of the rate of return. Journal of

Business 53.

Pesaran, M., Shin, Y., Smith, R., 2001. Bounds testing approaches to the analysis of level relationships.

Journal of Applied Econometrics 16, 289326.

Rogers, L., Satchell, S., 1991. Estimating variance from high, low, and closing prices. Annals of Applied

Probability.

Toda, H., Yamamoto, T., 1995. Statistical inferences in vector autoregressions with possibly integrated

processes. Journal of Econometrics 66, 225250.

Urquhrat, A., November 2016. The inefficiency of Bitcoin. Economics Letters 148, 8082.

Velde, F. R., December 2013. Bitcoin: A primer. Chicago Fed Letter (317).

White, L., 2015. The Market for Cyryptocurrencies. Cato Journal 35 (2).

Wilson, M., Yelowitz, A., September 2015. Characteristics of Bitcoin users: An analysis of Google search

data. Applied Economics Letters 22, 10301036.

Yermack, D., 2015. Is Bitcoin a real currency? An economic appraisal. In: Chuen, D. L. K. (Ed.), Handbook

of Digital Currency. Elsevier, pp. 3144.

24
Appendices

A Implications for bitcoin research

A.1 Are differences in bitcoin exchange prices sufficient to impact research?

The failure of some exchanges to achieve LOOP has important ramifications for those studying bitcoin. By

design, there is no central official source for bitcoin prices. Studies prior to February 2014 used the bitcoin

marketplace Mt. Gox for price information, as it handled approximately 70% of the global bitcoin trade.

Mt. Gox, however, declared bankruptcy in February 2014, and the largest exchange during the period we

study (BitFinex) handles only 11% of bitcoin transactions. As the sample of published bitcoin papers in

Table 12 illustrates, researchers currently use prices from any source that suits the purpose of their studyor

use prices from a bitcoin indexunder the assumption that their results generalize across all bitcoin markets.

Table 12: Sources of Bitcoin Price Information for papers after the collapse of Mt. Gox

Author Bitstamp BTC-e BitFinex BTC China Quandl/BPI/Index OKCoin

Badev and Chen (2014) X X X X X
Ciaian, Rajcaniova, and dArtis Kancs (2016) X
Chu, Nadarajah, and Chan (2015) X
Lo and Wang (2014) X
Urquhrat (2016) X

Might different research results be achieved by using a price-series from a different exchange? To inves-

tigate the extent of possible distortions we examine bitcoin volatility, a transparent statistic, to see if there

are substantial differences across exchanges and price-series. Table 13 lists the median volatility of a 30-day

rolling window of bitcoin prices, with exchanges grouped by adherence to LOOP. There are three different

measures of volatility employed for each market: the standard-deviation-based volatility based on a single

price-series, SD,x ; the Parkinson (1980) volatility, HL , based on High and Low prices; and the Rogers and

Satchell (1991) volatility, OHLC which uses Open, High, Low, and Close prices.

q qP
1 T 2
T t=1 (pxt px )
SD,x = (7)
px
q q PT
2
1 1 H
T 4 ln 2 t=1 pt pL
t
HL
= 1 H L
(8)
2 (p + p )
q qP
1 N


L

T t=1 pH C
t pt pH O + pL pC
t pt t t pt pO
t
OHLC = 1 (9)
4 (pO + pH + pL + pC )

25
 Table 13: Median 30-Day Price Volatility (%)

SD,x
Open Close Weighted Low High HL OHLC
LOOP Exchanges
ANXBTC 6.32 6.39 6.28 6.83 6.38 4.76 5.60
BitFinex 6.37 6.37 6.32 6.60 6.29 3.49 3.64
BitStamp 6.33 6.34 6.27 6.61 6.18 3.43 3.52
BPI 6.40 6.40 6.60 6.27 3.11 3.07
itBit 6.27 6.29 6.24 6.45 6.16 3.00 3.85
Kraken 6.40 6.45 6.32 6.61 6.27 3.79 4.09
LOOP Average 6.35 6.37 6.29 6.62 6.26 3.60 3.96

Weakly LOOP Failure Exchanges

LakeBTC 6.27 6.28 6.17 6.42 6.18 3.25 3.27
BitKonan 6.32* 6.93 6.49 7.08 6.98 5.30 6.37*

Strong LOOP Failure Exchanges

BTC-e 6.22 6.28 6.13 6.38 6.14 3.42 3.57
The Rock Trading 6.22 6.38 5.99 6.27* 6.51 4.78* 5.84*

Consistent LOOP Failure Exchanges

HitBTC 6.08 6.09 5.87 6.20 6.09 3.56 3.85
LocalBitcoins 20.37 19.08 5.97* 10.32 39.32 86.34 173.77
*:Obtained using the price series that cannot be tested for LOOP

Where x indicates the Open (O), High (H), Low (L), or Close (C) price respectively, px indicates the

average price, and exchange specific indexes are suppressed. To make the measures comparable across each

exchange and price series, each is normalized using the mean of the daily price for the period (or averaged

price if multiple price series are used). Over the same time period, the equivalent standard deviation-based

volatility of the price of gold is 1.8%, the USD-EUR exchange rate is 1.4%, and the S&P 500 is 1.3%.32

The data in Table 13 confirm that bitcoin is more volatile than gold, exchange rate, and stock market

prices, regardless of the exchange or price-serieshow much more volatile bitcoin is depends on the data

used. The volatility on LocalBitcoins is magnitudes larger than other exchanges for all price-series except

transaction-weighted prices. HitBTC yields the opposite: it usually has the lowest volatility. These results

show that using data from an exchange that fails to yield LOOP can impact empirical conclusions, but that

it is difficult to precisely predict the effect: The volatility of LocalBitcoin is consistently greater than that of

LOOP series, while the volatility of HitBTC is almost always below average. The exchanges that failed to

yield LOOP in the robustness checks are less consistent in their deviations from LOOP average than HitBTC

and LocalBitcoins.

Volatility varies even among LOOP exchanges. Volatility estimates using standard deviations vary from
32 We obtain the daily 10 a.m. Gold Fixing Price in London Bullion Market in USD, the USD-EUR exchange rate, and the

S&P 500 Index from the Federal Reserve Economic Database (FRED) of the St. Louis Federal Reserve Bank for the time
period of the bitcoin sample.

26
6.16% to 6.83% (a range of 0.67 percentage points). Estimates that combine different price-series, HL and

OHLC , vary even more. Parkinson volatility ranges from 3.00% to 4.76% (a 1.76 percentage point spread)

while the Rogers-Satchell volatility ranges from 3.07% to 5.60% (a 2.53 percentage point spread). Volatility

across bitcoin markets are highly dependent upon which exchange, price-series, and measure is used.

A.2 Correlations

Bitcoin prices are non-stationary, which biases correlation results. Despite this, some studies have drawn their

conclusions from only correlations. We repeat our analysis from section 4 but examine only the correlations

of the twelve exchanges to BitFinex (the largest exchange), with results presented in Table 14.

Table 14: Correlation Characteristics Across Exchanges

ANX BitK BitS BTC-e HitB itBi Krak Lake Local Rock
Open 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.87 1.00
Close 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.89 1.00
Weighted 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Low 1.00 0.99 1.00 1.00 1.00 0.98 1.00 1.00 0.97 1.00
High 1.00 0.99 1.00 1.00 1.00 1.00 1.00 1.00 0.14 0.99

The correlation results are similar regardless of which reference series is used and are typically close

to one, implying (incorrectly) that all series are interchangeable. Possible exceptions exist only for some

LocalBitcoins price-series. This contradiction of the cointegration results in section 4.2 should give pause

when basing interpretations on correlation results alone.

27
B Online Appendix: Additional Figures and Robustnesss Checks

B.1 Prices Levels

Figures (3) - (7) plot all price series for all exchanges. The two vertical lines denote the end of the first

segment and beginning of the second segment used for the structural break test in section 4.3.2. All price

series for all exchanges are still descending at the end of the first segment; all have plateaued by the beginning

of the second segment. The volatility of the Transaction Weighted price series for LocalBitcoins in Figure 5

is noticeably lower than it is for the other four price series.

Figure 3: Open Prices

28
 Figure 4: Close Prices

Figure 5: Weighted Prices

29
Figure 6: Low Prices

Figure 7: High Prices

30
B.2 Differences between exchange prices and BitFinex

Figures (8)-(12) plot the difference between the price-series of the indicated exchange and BitFinex, pit

pBitF
t
inex
. A trend line constructed with a Lowess smoother is included. The two vertical lines denote the

end of the first segment and beginning of the second segment for the structural break exercise in section

4.3.2.

Figure 8: Difference in Open Prices

31
 Figure 9: Difference in Close Prices

Figure 10: Difference in Weighted Prices

32
Figure 11: Difference in Low Prices

Figure 12: Difference in High Prices

33
B.3 Cointegration results, all series

For completeness, tables (15) - (17) contain the results of cointegration tests for all exchanges relative to all

other exchanges, using the methods of section 4.1. If a series does not affirm unit root, a result of N GC

indicates a single direction of granger-causality could not be established. An N C indicates that cointegration

was not found, while ` > 15 indicates that the series residual displayed auto-correlation up to lag 15. If none

of these issues were found, the cointegration parameter, is reported. All results are significant at 1%.

The results relative to BitFinex (the first column) are the results reported in section 4.2. LOOP status is

transitive: any series that consistently achieves LOOP with BitFinex has LOOP with any other series that

does the same. As documented in section 4.3.1, if a series does not achieve LOOP with BitFinex it does not

achieve LOOP with any other series.

Table 15: Cointegration with All Series, Open Prices

BitF ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock
ANXBTC 1.00
BitKonan ` > 15 0.98
BitStamp 1.00 1.00 ` > 15
BTC-e 0.99 0.99 ` > 15 0.99
HitBTC NC NC ` > 15 N C NC
itBit 1.00 1.00 ` > 15 1.00 1.01 NC
Kraken 1.01 1.00 1.02 1.00 1.02 NC 1.01
Lake 1.00 1.00 ` > 15 1.00 1.01 NC 1.00 0.99
Local 1.14 1.14 N GC 1.14 1.16 1.21 1.14 1.13 1.14
Rock 0.98 0.98 1.00 0.98 0.99 NC 0.99 0.98 0.99 1.15
BPI 1.00 0.99 ` > 15 1.00 1.01 NC 1.00 0.99 1.00 1.15 ` > 15
Abbreviations: ` > 15, Lag to avoid residual autocorrelation greater than 15 days; NC, Not Cointegrated.
NGC, Could not establish single direction granger-causality.

Table 16: Cointegration with All Series, Close Prices

BitF ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock
ANXBTC 1.01
BitKonan 0.98 0.98
BitStamp 1.00 0.99 1.01
BTC-e 0.99 0.98 1.01 0.99
HitBTC NC NC NC NC NC
itBit 1.00 0.99 1.01 1.00 1.01 NC
Kraken 1.00 1.00 1.02 1.00 1.01 NC 1.00
Lake 1.00 0.99 1.02 1.00 1.01 NC 1.01 1.00
Local 1.13 1.13 1.13 1.13 1.14 1.20 1.13 1.14 1.13
Rock 0.99 0.98 1.00 0.99 1.00 NC 0.99 0.98 0.99 1.15
BPI 1.00 0.99 1.01 1.00 1.01 NC 1.00 0.99 1.00 1.14 1.00
Abbreviations: NC, Not Cointegrated.

34
 Table 17: Cointegration with All Series, Weighted Prices

BitF ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local
ANXBTC 1.00
BitKonan 0.99 0.99
BitStamp 1.00 0.99 1.01
BTC-e 0.99 0.98 1.00 0.99
HitBTC NC NC NC NC NC
itBit 1.00 0.99 1.01 1.00 1.01 NC
Kraken 1.00 1.00 1.01 1.01 1.02 NC 1.01
Lake 1.00 0.99 1.01 1.00 1.01 NC 1.00 1.00
Local ` > 15 ` > 15 1.08 ` > 15 ` > 15 ` > 15 ` > 15 1.07 ` > 15
Rock 0.98 0.98 1.00 0.99 0.99 NC 0.98 0.98 0.98 0.92
Abbreviations: ` > 15, Lag to avoid residual autocorrelation greater than 15 days; NC, Not Cointegrated.

Table 18: Cointegration with All Series, Low Prices

BitF ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock
ANXBTC 0.98
BitKonan 0.98 0.99
BitStamp 1.00 1.01 1.02
BTC-e 0.99 1.00 1.02 0.99
HitBTC NC 0.95 NC NC NC
itBit 1.00 1.02 1.03 1.01 1.01 NC
Kraken 1.00 1.03 1.03 1.01 1.02 1.09 1.01
Lake 1.00 1.02 1.02 1.01 1.01 NC 1.00 1.00
Local 0.89 0.91 0.91 0.90 0.91 0.95 0.90 0.89 0.89
Rock ` > 15 1.01 1.02 ` > 15 ` > 15 ` > 15 0.99 0.98 0.99 0.90
BPI 1.00 1.01 1.02 1.00 1.01 NC 0.99 0.99 0.99 0.90 ` > 15
Abbreviations: ` > 15, Lag to avoid residual autocorrelation greater than 15 days; NC, Not Cointegrated.

Table 19: Cointegration with All Series, High Prices

BitF ANX BitK BitS BTC-e HitBTC itBit Kraken Lake Local Rock
ANXBTC 1.02
BitKonan 0.99 0.98
BitStamp 1.00 0.98 1.01
BTC-e 0.98 0.97 0.99 0.99
HitBTC NC NC 0.95 NC NC
itBit 0.99 0.97 1.01 1.00 1.01 1.07
Kraken 1.00 0.99 1.01 1.00 1.02 NC 1.01
Lake 1.00 0.98 1.00 1.00 1.01 NC 1.00 0.99
Local 1.06 NGC 1.03 1.07 NGC NGC NGC NGC NGC
Rock 0.97 0.95 0.98 0.97 0.98 NC 0.97 0.96 0.97 NGC
BPI 0.99 0.98 1.00 0.99 1.01 NC 1.00 0.99 1.00 NGC 1.03
Abbreviations:NC, Not Cointegrated. NGC, Could not establish single direction granger-causality.

B.4 Price Deviation Results

Tables 20 - 37 represent variations of the data and regressions presented in section 5.3. Results show a

decrease in price deviations associated with a requirement of ID for funding.

B.4.1 Summary Statistics

Table 20 is an expansion of Table 9, including the Min and Max values for each exchange.

35
 Table 20: Detailed Summary Statistics, Price Deviation and Trade Volume

market Open Close Weighted High Low TVR

ANXBTC Mean -5.47 -5.32 -5.08 -4.99 -4.88 0.08
Std. Dev 1.38 1.31 1.22 1.31 1.44 0.10
Min -9.88 -10.14 -10.45 -9.53 -9.83 0.00
Max -2.86 -2.47 -2.72 -1.65 -1.64 0.80
BitKonan Mean -4.35 -4.09 -4.23 -4.09 -4.16 0.00
Std. Dev 1.06 1.11 1.17 1.31 1.19 0.00
Min -10.22 -8.15 -9.46 -8.85 -10.13 0.00
Max -1.92 -1.68 -1.94 -1.00 -0.99 0.01
BitStamp Mean -6.08 -6.07 -6.00 -5.77 -5.75 0.63
Std. Dev 1.26 1.27 1.20 1.24 1.27 0.38
Min -10.52 -10.66 -10.50 -10.09 -10.55 0.17
Max -2.39 -2.84 -3.33 -3.11 -2.28 3.31
BTC-e Mean -4.59 -4.64 -4.60 -4.50 -4.64 0.41
Std. Dev 0.93 1.00 0.93 0.87 1.06 0.24
Min -10.07 -9.38 -9.78 -9.39 -10.47 0.02
Max -2.46 -2.56 -2.65 -1.84 -2.42 2.88
HitBTC Mean -4.05 -3.99 -4.01 -4.10 -3.93 0.04
Std. Dev 1.64 1.59 1.59 1.77 1.52 0.06
Min -9.39 -9.15 -8.97 -10.98 -9.49 0.00
Max -1.52 -1.56 -1.59 -1.36 -1.63 0.44
itBit Mean -5.70 -5.72 -5.44 -5.23 -5.52 0.16
Std. Dev. 1.18 1.20 1.20 1.04 1.36 0.17
Min -11.10 -10.95 -10.10 -10.98 -10.52 0.00
Max -3.35 -3.26 -2.55 -2.76 -2.16 1.43
Kraken Mean -4.85 -4.77 -5.02 -4.79 -4.92 0.00
Std. Dev. 1.36 1.25 1.36 1.24 1.26 0.00
Min -11.05 -10.29 -10.39 -9.08 -9.03 0.00
Max -1.67 -1.69 -1.89 -1.89 -1.35 0.01
LakeBTC Mean -5.41 -5.39 -5.39 -5.32 -5.17 0.30
Std. Dev 1.05 1.00 1.07 1.08 1.05 0.27
Min -10.16 -9.37 -10.14 -10.69 -9.27 0.02
Max -3.08 -3.15 -3.07 -3.29 -2.31 2.92
LocalBitcoins Mean -2.02 -2.09 -2.24 0.78 -2.06 0.10
Std. Dev 1.17 1.14 0.28 0.65 0.50 0.09
Min -9.66 -6.92 -3.14 -0.41 -5.27 0.00
Max 0.64 0.55 -1.37 3.15 -1.01 0.51
The Rock Trading Mean -4.51 -4.61 -4.56 -4.14 -4.71 0.00
Std. Dev 1.12 1.17 1.07 1.28 1.34 0.00
Min -9.57 -10.20 -9.93 -9.89 -9.15 0.00
Max -2.04 -2.00 -1.71 -0.42 -2.01 0.01

B.4.2 Single Exchange Characteristic

Tables (21)-(25) report single effect results from the OLS regression on price deviation (regression (i) in Table

11). Results of statistically significant variables usually retain the same sign regardless of which price-series

is used. There are only two exceptions, both of which occur for the High price: regression (viii) (Fixed

Trading Fee) and regression (ix) (Indicator for Low Global Share), which increase deviations for all except

36
the High price-series.

Table 21: Price Deviation and Single Exchange Characteristics, OPEN price-series
Exchange Transparency Regulations Fees Exchange Size Other Exchange Characteristics
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii) (xiv) (xv) (xvi) (xvii)

Know Incorporation 0.02

know Physical Location -0.48***
Know Server Location -1.17***
USA Location -1.21***
ID to Fund -1.38***
ID for Fiat Withdrawal -1.02***
Free Fiat Deposit -0.61***
Fixed Trading Fee 0.15**
Free Trades Feasible 0.59***
High Global Share -0.79***
Low Global Share 0.18***
In BPI -0.79***
Trade EUR 0.22***
Trade GBP 0.23***
Trade non-EUR/GBP 0.08
Hacked 0.81***
Offers Debit Card -1.34***
Constant -4.72*** -4.32*** -4.23*** -4.34*** -4.15*** -4.08*** -4.52*** -4.76*** -4.88*** -4.54*** -4.76*** -4.54*** -4.88*** -4.82*** -4.75*** -4.86*** -4.43***

Date fixed effect? Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

Adj. R2 -0.04 -0.03 0.09 0.08 0.15 0.06 -0.01 -0.04 -0.01 -0.00 -0.04 -0.00 -0.04 -0.04 -0.04 0.00 0.08
N 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243
Significance: 1%, **5%, *10% .

Table 22: Price Deviation and Single Exchange Characteristics, CLOSE price-series
Exchange Transparency Regulations Fees Exchange Size Other Exchange Characteristics
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii) (xiv) (xv) (xvi) (xvii)

Know Incorporation 0.02

Know Physical Location -0.44***
Know Server -1.06***
USA Location -1.22***
ID to Fund -1.34***
ID for Fiat Withdrawal -1.03***
Free Fiat Deposit -0.62***
Fixed Trading Fee 0.26* **
Free Trades Feasible 0.58***
High Global Share -0.86***
Low Global Share 0.25***
In BPI -0.86***
Trade EUR 0.08
Trade GBP 0.19***
Trade non-EUR/GBP 0.04
Hacked 0.73***
Offers Debit Card -1.28***

Constant -4.68*** -4.31*** -4.24*** -4.30*** -4.13*** -4.04*** -4.48*** -4.77* ** -4.84*** -4.49*** -4.74*** -4.49*** -4.73*** -4.76*** -4.69*** -4.81*** -4.41***
Date fixed effect? Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Adj. R2 -0.03 -0.02 0.08 0.10 0.15 0.08 0.00 -0.03 -0.00 0.02 -0.03 0.02 -0.03 -0.03 -0.03 0.00 0.08
N 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243 4243
Significance: 1%, **5%, *10% .

Table 23: Price Deviation and Single Exchange Characteristics, WEIGHTED price-series
Exchange Transparency Regulations Fees Exchange Size Other Exchange Characteristics
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii) (xiv) (xv) (xvi) (xvii)

Know Incorporation 0.04

Know Physical Location -0.43***
Know Server Location -0.89***
USA Location -1.19***
ID to Fund -1.22***
ID for Fiat Withdrawal -0.91***
Free Fiat Deposit -0.50***
Fixed Trading Fee 0.29***
Free Trades Feasible 0.61***
High Global Share -0.81***
Low Global Share 0.07
In BPI -0.81***
Trade EUR 0.18***
Trade GBP 0.15**
Trade Non-EUR/GBP 0.08
Hacked 0.68***
Offers Debit Card -1.10***
Constant -4.69*** -4.31*** -4.30*** -4.30*** -4.17*** -4.11*** -4.50*** -4.77*** -4.84*** -4.49*** -4.67*** -4.49*** -4.80*** -4.73*** -4.70*** -4.79*** -4.43***

Date fixed effect? Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

Adj. R2 -0.02 -0.01 0.07 0.12 0.15 0.07 0.00 -0.01 0.01 0.03 -0.02 0.03 -0.02 -0.02 -0.02 0.01 0.07
N 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238 4238
Significance: 1%, **5%, *10% .

37
 Table 24: Price Deviation and Single Exchange Characteristics, HIGH price-series
Exchange Transparency Regulations Fees Exchange Size Other Exchange Characteristics
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii) (xiv) (xv) (xvi) (xvii)

Know Incorporation 0.62***

Know Physical Location 0.11
Know Server Location -1.34***
USA Location -1.50***
ID to Fund -1.64***
ID for Fiat Withdrawal -1.57***
Free Fiat Deposit -0.86***
Fixed Trading Fee -0.35***
Free Trades Feasible 1.50***
High Global Share -1.15***
Low Global Share -0.18**
In BPI -1.15***
Trade EUR 0.61***
Trade GBP 0.86***
Trade Non-EUR/GBP 0.52***
Hacked 2.16***
Offers Debit Card -1.46***
Constant -4.71*** -4.30*** -3.67*** -3.76*** -3.55*** -3.27*** -3.95*** -4.07*** -4.66*** -3.98*** -4.15*** -3.98*** -4.70*** -4.64*** -4.52*** -4.64*** -3.92***

Date fixed effect? Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

Adj. R2 -0.05 -0.07 0.04 0.05 0.09 0.08 -0.03 -0.06 0.05 -0.02 -0.07 -0.02 -0.05 -0.02 -0.05 0.12 0.02
N 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241 4241
Significance: 1%, **5%, *10% .

Table 25: Price Deviation and Single Exchange Characteristics, LOW price-series
Exchange Transparency Regulations Fees Exchange Size Other Exchange Characteristics
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii) (xiv) (xv) (xvi) (xvii)

Know Incorpation -0.03

Know Physical Location -0.35***
Know Server Location -0.84***
USA Location -1.18***
ID to Fund -1.16***
ID for Fiat Withdrawal -0.95***
Free Fiat Deposit -0.49***
Fixed Trading Fee 0.28***
Free Trading Fee 0.78***
High Global Share -0.78***
Low Global Share -0.04
In BPI -0.78***
EUR Trade 0.11*
GBP Trader 0.22***
Trade Non-EUR/GBP 0.11*
Hacked 0.85***
Offers Debit Card -0.93***
Constant -4.54*** -4.29*** -4.23*** -4.21*** -4.10*** -4.00*** -4.42*** -4.68*** -4.80*** -4.41*** -4.56*** -4.41*** -4.66*** -4.68*** -4.63*** -4.74*** -4.38***

Date fixed effect? Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

Adj. R2 -0.01 -0.00 0.06 0.12 0.13 0.08 0.01 -0.01 0.04 0.03 -0.01 0.03 -0.01 -0.01 -0.01 0.04 0.05
N 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233 4233
Significance: 1%, **5%, *10% .

B.4.3 Combined Exchange Characteristics

Tables (26)-(30) examines grouped regressions of exchange characteristics for groupings that do not result

in multicollinearity. The interaction of Fixed Cost to Trade and Free Trades Feasible is a fixed effect for

ANXBTC, and is not used. Increases in transparency (Known Server Location) decrease deviations, as does

requiring an ID to Fund or having a larger exchange. Fees increase deviations. Accepting trades in multiple

foreign currencies has a mixed effect, though overall an exchange that allows for trade in Euros, GBP, and

at least one other currency has larger deviations than an exchange that does not.

38
Table 26: Price Deviation and Grouped Exchange Characteristics, OPEN price-series
Transparency Regulations Fees Exchange Size Other
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii)

Know Server Location -0.88*** -0.80***

USA Location -0.87*** -0.75***
Server x USA -0.23*

ID to Fund -1.38***

Fixed Trade Cost 0.19***

Free Trade Feasible 0.60***

Low Global Share -0.21*** -0.08

Trade Volume -1.62*** -1.64***
Low x Volume -186.64***

EUR Trade 0.09

GBP Trade 0.19** -1.10***
Non EUR/GBP Trade 0.07 -1.03***
GBPxNon EUR/GBP 2.22***

Constant -4.09*** -4.11*** -4.15*** -4.96*** -4.36*** -4.35*** -4.91*** -4.20***

Date fixed effect? Y Y Y Y Y Y Y Y
Adj. R2 0.15 0.15 0.15 -0.01 0.01 0.02 -0.04 0.08
N 4243 4243 4243 4243 4243 4243 4243 4243
Significance: 1%, **5%, *10% .

Table 27: Price Deviation and Grouped Exchange Characteristics, CLOSE price-series
Transparency Regulations Fees Exchange Size Other
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii)

Know Server Location -0.74*** -0.56***

USA Location -0.94*** -0.63***
Server x USA -0.56***

ID to Fund -1.34***

Fixed Trade Cost 0.31***

Free Trades Feasible 0.61***

Low Global Share -0.15** 0.05

Trade Volume -1.66*** -1.70***
Low Share x Volume -284.83***

Trade EUR -0.06

Trade GBP 0.21*** -1.30***
Trade Non-EUR/GBP 0.04 -1.21***
GBP x Non-EUR/GBP 2.49***

Constant -4.09*** -4.14*** -4.13*** -4.97*** -4.33*** -4.32*** -4.75*** -4.04***

Date fixed effect? Y Y Y Y Y Y Y Y
2
Adj. R 0.15 0.15 0.15 0.01 0.03 0.04 -0.03 0.13
N 4243 4243 4243 4243 4243 4243 4243 4243
Significance: 1%, **5%, *10% .

39
Table 28: Price Deviation and Grouped Exchange Characteristics, WEIGHTED price-series
Transparency Regulations Fees Exchange Size Other
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii)

Know Server Location -0.57*** -0.50***

USA Location -0.97*** -0.87***
Server x USA -0.20*

ID to Fund -1.22***

Fixed Trade Costs 0.34***

Free Trade Feasible 0.65***

Low Global Share -0.35*** -0.20***

Trade Volume -1.70*** -1.73***
Low Share x Volume -214.34***

Trade EUR 0.11

Trade GBP 0.11* -1.16***
Trade Non-EUR/GBP 0.07 -1.02***
GBP x Non-EUR/GBP 2.18***

Constant -4.13*** -4.15*** -4.17*** -4.98*** -4.26*** -4.25*** -4.83*** -4.12***

Date fixed effect? Y Y Y Y Y Y Y Y
2
Adj. R 0.15 0.15 0.15 0.03 0.04 0.05 -0.02 0.12
N 4238 4238 4238 4238 4238 4238 4238 4238

Significance: 1%, **5%, *10% .

Table 29: Price Deviation and Grouped Exchange Characteristics, HIGH price-series
Transparency Regulations Fees Exchange Size Other
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii)

Know Server Location -0.97*** -1.09***

USA Location -1.13*** -1.34***
Server x USA 0.39***

ID to Fund -1.64***

Fixed Trade Costs -0.23***

Free Trade Feasible 1.47***

Low Global Share -0.79*** -0.70***

Trade Volume -2.47*** -2.49***
Low Share x Volume -130.52**

Trade EUR 0.00

GBP Trade 0.86*** -0.86***
Trade Non-EUR/GBP 0.52*** -0.92***
GBP x Non-EUR/GBP 2.87***

Constant -3.48*** -3.45*** -3.55*** -4.56*** -3.55*** -3.54*** -4.95*** -4.09***

Date fixed effect? Y Y Y Y Y Y Y Y
2
Adj. R 0.10 0.10 0.09 0.05 0.01 0.01 -0.01 0.11
N 4241 4241 4241 4241 4241 4241 4241 4241
Significance: 1%, **5%, *10% .

40
B.4.4 Grouped Exchange Characteristics

Tables (31)-(35) moves iteratively from regression (i) to regression (ii) in Table 11. The result that ID to

Fund decreases deviations never changes, though the magnitude may vary.

41
Table 30: Price Deviation and Grouped Exchange Characteristics, LOW price-series
Transparency Regulations Fees Exchange Size Other
(i) (ii) (iii) (iv) (v) (vi) (vii) (viii)

Know Server Location -0.52*** -0.42***

USA Location -0.98*** -0.83***
Know Server x USA -0.29**

ID to Fund -1.16***

Fixed Trade Cost 0.35***

Free Trade Feasible 0.81***

Low Global Share -0.44*** -0.25***

Trade Volume -1.65*** -1.69***
Low Share x Volume -274.65***

Trade EUR -0.06

Trade GBP 0.25*** -1.18***
Trade Non-EUR/GBP 0.11* -1.06***
GBP x Non-EUR/GBP 2.34***

Constant -4.07*** -4.09*** -4.10*** -4.95*** -4.15*** -4.14*** -4.71*** -4.05***

Date fixed effect? Y Y Y Y Y Y Y Y
Adj. R2 0.14 0.14 0.13 0.06 0.04 0.05 -0.01 0.14
N 4233 4233 4233 4233 4233 4233 4233 4233
Significance: 1%, **5%, *10% .

Table 31: Price Deviation and Exchange Characteristics, OPEN price-series

(i) (ii) (iii) (iv) (v) (vi)

ID to Fund -1.38*** -1.34*** -1.66*** -1.33*** -1.44*** -1.92***

ID to Fund x Low Share 1.22*** 1.50***

Fixed Cost to Trade -0.12* -0.04 -0.24*** -0.32*** -0.12*

Free Trade Feasible 0.58*** 0.54*** 0.38*** 0.78*** 0.83***
Low Global Share 0.29*** -0.16** -0.09 0.06 -0.40***

Trade Volume -1.16*** -0.77*** -0.33***

Volume x Fixed Cost -0.08 -1.02***

Volume x Free Trade -2.42*** -3.06***
Volume x Low Share -94.95** -167.55***

Constant -4.15*** -4.38*** -4.25*** -3.96*** -3.97*** -3.87***

Date fixed effect? Y Y Y Y Y Y
Adj. R2 0.15 0.17 0.20 0.19 0.21 0.25
N 4243 4243 4243 4243 4243 4243
Significance: 1%, **5%, *10% .

42
 Table 32: Price Deviation and Exchange Characteristics, CLOSE price-series

(i) (ii) (iii) (iv) (v) (vi)

ID to Fund -1.34*** -1.24*** -1.56*** -1.23*** -1.32*** -1.79***

ID to Fund x Low Share 1.20*** 1.50***

Fixed Cost to Trade 0.04 0.12* -0.08 -0.15** 0.05

Free Trade Feasible 0.65*** 0.61*** 0.46*** 0.78*** 0.82***
Low Global Share 0.41*** -0.03 0.05 0.23*** -0.23**

Trade Volume -1.11*** -0.82*** -0.39***

Volume x Fixed Cost -0.13 -1.07***

Volume x Free Trade -1.92*** -2.56***
Volume x Low Share -175.80*** -248.08***

Constant -4.13*** -4.50*** -4.38*** -4.10*** -4.10*** -3.99***

Date fixed effect? Y Y Y Y Y Y
Adj. R2 0.15 0.18 0.21 0.20 0.22 0.25
N 4243 4243 4243 4243 4243 4243
Significance: 1%, **5%, *10% .

Table 33: Price Deviation and Exchange Characteristics, WEIGHTED price-series

(i) (ii) (iii) (iv) (v) (vi)

ID to Fund -1.22*** -1.13*** -1.34*** -1.12*** -1.22*** -1.53***

ID to Fund x Low Share 0.78*** 0.98***

Fixed Cost to Trade 0.07 0.13** -0.05 -0.12* 0.01

Free Trade Feasible 0.62*** 0.59*** 0.42*** 0.75*** 0.78***
Low Global Share 0.23*** -0.06 -0.16** -0.02 -0.32***

Trade Volume -1.20*** -0.87*** -0.58***

Volume x Fixed Cost -0.12 -0.73***

Volume x Free Trade -2.05*** -2.47***
Volume x Low Share -112.55** -159.61***

Constant -4.17*** -4.48*** -4.41*** -4.05*** -4.06*** -3.99***

Date fixed effect? Y Y Y Y Y Y
Adj. R2 0.15 0.18 0.19 0.21 0.22 0.24
N 4238 4238 4238 4238 4238 4238
Significance: 1%, **5%, *10% .

43
 Table 34: Price Deviation and Exchange Characteristics, HIGH price-series

(i) (ii) (iii) (iv) (v) (vi)

ID to Fund -1.64*** -1.68*** -1.94*** -1.67*** -1.88*** -2.27***

ID to Fund x Low Share 0.97*** 1.25***

Fixed Cost to Trade -0.65*** -0.58*** -0.84*** -1.00*** -0.83***

Free Trade Feasible 1.36*** 1.33*** 1.05*** 1.82*** 1.86***
Low Global Share 0.17*** -0.18** -0.42*** -0.19* -0.57***

Trade Volume -1.82*** -1.11*** -0.74***

Volume x Fixed Cost 0.05 -0.74***

Volume x Free Trade -4.53*** -5.07***
Volume x Low Share -99.56* -160.24***

Constant -3.55*** -3.73*** -3.64*** -3.08*** -3.12*** -3.03***

Date fixed effect? Y Y Y Y Y Y
Adj. R2 0.09 0.21 0.22 0.24 0.28 0.30
N 4241 4241 4241 4241 4241 4241
Significance: 1%, **5%, *10% .

Table 35: Price Deviation and Exchange Characteristics, LOW price-series

(i) (ii) (iii) (iv) (v) (vi)

ID to Fund -1.16*** -1.06*** -1.29*** -1.05*** -1.16*** -1.53***

ID to Fund x Low Share 0.88*** 1.18***

Fixed Cost to Trade 0.09* 0.15** -0.02 -0.07 0.09

Free Trade Feasible 0.78*** 0.75*** 0.60*** 0.96*** 0.99***
Low Global Share 0.20*** -0.12* -0.14* 0.04 -0.32***

Trade Volume -1.06*** -0.67*** -0.33**

Volume x Fixed Cost -0.32* -1.07***

Volume x Free Trade -2.21*** -2.72***
Volume x Low Share -169.98*** -227.45***

Constant -4.10*** -4.48*** -4.39*** -4.10*** -4.10*** -4.01***

Date fixed effect? Y Y Y Y Y Y
Adj. R2 0.13 0.18 0.19 0.20 0.22 0.24
N 4233 4233 4233 4233 4233 4233
Significance: 1%, **5%, *10% .

44
B.4.5 Controlling for possible autocorrelation in the residual

Tables (36) and (37) reports the result from the OLS regression using Newey-West errors, allowing for

autocorrelation in the residual up to lag length `. A Newey West with lag 0 is the same as robust standard

errors reported in Table 11 (ii). All results regarding the ID to Fund remain statistically significant.

Table 36: Price Deviation and Exchange Characteristics, Regression (i), Newey-West Regression
Open Close Weighted High Low
`= 0 2 15 0 2 15 0 2 15 0 2 15 0 2 15

ID to Fund -1.38*** -1.38*** -1.38*** -1.34*** -1.34*** -1.34*** -1.22*** -1.22*** -1.22*** -1.64*** -1.64*** -1.64*** -1.16*** -1.16*** -1.16***

Constant -4.59*** -4.59*** -4.59*** -3.63*** -3.63*** -3.63*** -4.38*** -4.38*** -4.38*** -2.98*** -2.98*** -2.98*** -3.80*** -3.80*** -3.80***
Date fixed effect? Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
N 4243 4243 4243 4243 4243 4243 4238 4238 4238 4241 4241 4241 4233 4233 4233

Significance: 1%, **5%, *10% .

Table 37: Price Deviation and Exchange Characteristics, Regression (ii), Newey-West Regression
Open Close Weighted High Low
`= 0 2 15 0 2 15 0 2 15 0 2 15 0 2 15

ID to Fund -1.92*** -1.92*** -1.92*** -1.79*** -1.79*** -1.79*** -1.53*** -1.53*** -1.53*** -2.27*** -2.27*** -2.27*** -1.53*** -1.53*** -1.53***
Low Global Share -0.40*** -0.40*** -0.40 -0.23** -0.23* -0.23 -0.32*** -0.32** -0.32 -0.57*** -0.57*** -0.57* -0.32*** -0.32** -0.32
ID to Fund x Low Share 1.50*** 1.50*** 1.50*** 1.50*** 1.50*** 1.50*** 0.98*** 0.98*** 0.9 8*** 1.25*** 1.25*** 1.25*** 1.18*** 1.18*** 1.18***
Trade Volume -0.33*** -0.33** -0.33 -0.39*** -0.39** -0.39* -0.58*** -0.58*** -0.58** -0.74*** -0.74*** -0.74** -0.33** -0.33* -0.33
Fixed Trading Costs -0.12* -0.12 -0.12 0.05 0.05 0.05 0.01 0.01 0.01 -0.83*** -0.83*** -0.83*** 0.09 0.09 0.09
Volume x Fixed Trading -1.02*** -1.02*** -1.02* -1.07*** -1.07*** -1.07* -0.73*** -0.73** -0.73 -0.74*** -0.74* -0.74 -1.07*** -1.07*** -1.07**
Free Trade Feasible 0.83*** 0.83*** 0.83*** 0.82*** 0.82*** 0.82*** 0.78*** 0.78*** 0.78*** 1.86*** 1.86*** 1.86*** 0.99*** 0.99*** 0.99***
Volume x Free Trade -3.06*** -3.06*** -3.06*** -2.56*** -2.56*** -2.56*** -2.47*** -2.47*** -2. 47*** -5.07*** -5.07*** -5.07*** -2.72*** -2.72*** -2.72***
Volume x Low Share -167.55*** -167.55*** -167.55*** -248.08*** -248.08*** -248.08*** -159.61*** -159.61*** -159.61* ** -160.24*** -160.24** -160.24* -227.45*** -227.45*** -227.45***

Constant -4.43*** -4.43*** -4.43*** -3.61*** -3.61*** -3.61*** -4.31*** -4.31*** -4.31*** -2.67*** -2.67*** -2.67*** -3.83*** -3.83*** -3.83***
N 4243 4243 4243 4243 4243 4243 4238 4238 4238 4241 4241 4241 4233 4233 4233

Significance: 1%, **5%, *10% .

45