Bitcoin and the Mt. Gox crisis Student: Pepijn van der Wal Student number: 5940370 Supervisor: M.A. Dijkstra Msc Date: 18 July 2014
Abstract:
When it eventually collapsed, Mt. Gox was the oldest bitcoin exchange still in operation. Once the near monopolist of the US dollar to Bitcoin exchange market with a market share of nearly 85% in January 2013; dollar and bitcoin withdrawal problems caused due to regulatory problems brought the exchange down. A study has been performed to determine the effect of the Mt. Gox collapse on the bitcoin exchange rate return. Beyond a higher spread between the Bitstamp and BTC-E exchanges during the final days of the Mt. Gox exchange, no evidence is found that suggests the demise of Mt. Gox had a significant effect on the bitcoin exchange price return.
Introduction
Satoshi Nakamoto published a paper in 2008 introducing Bitcoin to the world. In the paper Nakamoto proposes a so-called peer-to-peer currency. The proposed currency would be decentralised and would enable digital transactions without the interference of a financial institutions. (Nakamoto, 2008) The currency was launched in 2009 with the mining of the first 50 bitcoin.
Currently, 1 June 2014, over 1,283,000 bitcoins have been mined and at a market price of $674.98 per bitcoin the total value of these bitcoins is just over $ 8.291 billion. In May 2014, over 1,997 million separate transactions have been
performed using bitcoin as a currency. (Blockchain.info)
Bitcoin has created a new market. Specific hardware has been developed to mine bitcoins, payment providers have been set up to make accepting bitcoin easy, digital wallets act as the digital bitcoin version of banks, although they currently rarely pay interest or engage in fractional reserve banking and bitcoin exchange allow the user to buy and sell bitcoins for traditional currencies.
One of the biggest and oldest bitcoin exchanges was the Mt. Gox exchange. The name, Mt. Gox, is derived from its origins as online card trading platform (Magic the Gather online exchange), although it is best known for trading bitcoin. The exchange quickly became one of the most popular bitcoin exchanges and in January 2013 it had a market share of 84.6% in the US dollar to bitcoin exchange market.
Despite its apparent success, the exchange also faced problems. Former partner CoinLab sued the exchange; funds were seized by the Department of Homeland security; verification problems of credit deposits led to significant losses and problems regarding delays of US dollar withdrawals surfaced. The exchange finally halted all trading on February 24th and filed for bankruptcy protection
This research looks into the effect of the demise of one of the largest bitcoin focussed companies yet. To determine the effect of the failure of the Mt. Gox exchange on the bitcoin market, its effect on the price of bitcoin on the Mt. Gox exchange is examined as well as the effect on the price of bitcoin on the Bitstamp and the BTC-E exchange are examined. Additionally the spread between the exchange rate between Mt. Gox, Bitstamp and BTC-E are examined.
The research finds no proof that the downfall of Mt. Gox had a significant effect on the exchange rate returns. As expected, the exchange rate return rose when dollars could not be withdrawn (without delay) and exchange rate return plummeted when the bitcoin withdrawals halted as well. The effect on the
exchange price returns on the Bitstamp and BTC-E exchanges have been found to be, beyond a larger spread for the duration of the crisis, negligible.
In the next section an overview of the literature is presented after which a research question and hypotheses are formulated. The methodology section discusses the set-up of the research, discusses the regression including the dependant variable, the main independent variables and the control variables. A description of the dataset is given in the data section after which a discussion of the results follows. The paper ends with a conclusion.
Literature
Currencies have three main attributes according to economists; they are a medium of exchange, unit of account and a store of value. (IMF, 2000) Bitcoin performs weak in each of these three attributes. (Yermack, 2014)
Although their number is growing, only a few merchants accept bitcoin as a medium of exchange, and acceptance doesn’t come near the level of acceptance of traditional currencies. Yermack (2014) argues bitcoins are a weak unit of account due to the relatively high value of one bitcoin. A $2 packet of gum would cost 0.00309 bitcoin making it difficult for both user and merchant to use. The frequent hacks and thefts of bitcoins make bitcoin a weak store of value. Out of
the 40 exchanges examined by Moore and Christin (2013), 9 fell victim to hacks. Of these 9 exchanges 5 have closed during the window of Moore and Christins’ research and two more have closed since then (until June 1st, 2014).
If bitcoin is not a good currency according to the standards of the traditional currencies, what does make bitcoin an interesting phenomenon? Barber, Boyen, Shi and Uzun (2012) list a few interesting characteristics which make bitcoin stand out when compared to traditional currencies. The main characteristics being the lack of a central point of trust, the fixed rate at which the supply of money grows and the low transaction fees. Additionally to these characteristics there is interest in the built in anonymity of the transactions, especially by parties using it to launder money or buy/sell illicit goods and services. (Moser, Bohme, & Breuker, 2013)
In a traditional currency such as the United States dollar (USD) and the British Pound (GBP) the main central point of trust is the central bank. The central bank uses its power to ensure the health of the banking system, aims to provide price stability by controlling the money supply and interest rates. (Goodhart, 1995) Bitcoin has no such central point of trust, there is no one institution overseeing or controlling the bitcoin market. The majority of miners in effect control the bitcoin network. Bitcoin donate computing power to the bitcoin network to process payments, if one party controls more than 50% of the total computing power, the party could theoretically spend a bitcoin multiple times. Since they would be validating their own payment. Therefore, it can be said that the bitcoin network assumes the majority of its miners are honest (Barber et al., 2012)
An algorithm that ensures the amount of bitcoins in the market rises at a predictable level controls the money supply. Currently 25 bitcoins are mined approximately every 10 minutes, the amount of bitcoins mined per 10 minutes will halve every four years until the total supply of 21 million bitcoins is
approached. (Grinberg, 2012) Since there is no central bank and no fractional reserve banking, the money supply grows only with the amount of bitcoins that are mined according to the algorithm.
A year after mining the first bitcoins in 2009, interest in bitcoin remained fairly contained, with little interest beyond a select group of cryptographers. The group of early-adopters’ main interested was the idea of bitcoin as a digital cryptocurrency. On several occasions early adopters have purchased bitcoins to hand them out to spread the word, such as Gavin Andresen who gave away 10,000 bitcoins (Wallace, 2011), currently (1 June 2014) valued at
approximately $6,464,800.
Early price developments were caused mostly by increase in demand. After the establishment of the first Bitcoin exchange, BitcoinMarket, on 6 February, exposure on the Slashdot website on 11 July 2010 was followed by a 1000% increase in exchange price. The increased demand drove the price up from $0.008 per bitcoin to $0.080 per bitcoin. On 17 July Mt. Gox became the second active bitcoin exchange. (historyofbitcoin.org)
As the interest in bitcoin spread, the demand in bitcoins grew along with it. While only 5,043 bitcoins were traded in January 2011, this had increased with 589% to 34,736 in January 2012. The growth of the popularity of bitcoin was accompanied by the introduction of new services using and accepting bitcoins such as bitcoin exchanges, digital wallets and financial services. The rise in popularity and availability of (financial) services changed brought new users with them. Speculators were drawn to and were causing high fluctuations in price (European Central Bank, 2012) and other users were more interested in the anonymity of bitcoin and used bitcoin to launder money and purchase illicit goods, such as illegal narcotics. (Wallace, 2011)
Silk Road was one of the more notorious new services, an online auction house, which allowed its users to buy and sell goods most of them illegal drugs. In 2012 a monthly dollar equivalent of $ 1.22 million in goods were sold through the Silk Road site (Christin, 2013). Research by Moser, Bohme and Breuker (2013) found several money laundering services processing accounts worth more than a million dollar.
Currently the vast majority of literature suggests the main driving force behind bitcoin price development is driven by speculation. (Gomez-Gonzalez & Parra-Polania, 2014; Hanley, 2013; Kristoufek, 2013, 2014; MacDonell, 2014)
Speculators are only interested in future value of bitcoin, as bitcoin has no intrinsic value and since it pays no dividend or interest, the only way to make money as a speculator is an increase in the bitcoin exchange price.
Demirgüç-Kunt and Levine (1996) found evidence concerning the level of development of the financial intermediaries market and the volatility of that market. The bitcoin financial market can be seen as developing. Bitcoin hardly has a banking system. Financial services such as wallets and exchanges make up the most of the available services. Online wallets have been victim to hacks, most notably Inputs.io and BIPS both losing over $1 million in worth of bitcoins at the time of announcement (of the hack). (Hern, 2014) Bitcoin exchanges are hardly stable either considering their high failure rate. (Moore & Christin, 2013)
In 2010 Mt. Gox was launched as one of the first bitcoin exchanges, at the time Mt. Gox launched the only other active exchange was Bitcoin Market. The launch of exchanges made it easier to acquire and dispose of bitcoins.
Mt. Gox used to have the largest market share of the US Dollar to Bitcoin exchange market. In January 2013 the Mt. Gox exchange was by far the largest with a market share of 84.6% in the US Dollar to Bitcoin exchange market. In February 2014, the month Mt. Gox filed for bankruptcy protection in Japan, the market share of the exchange had dropped to 25.4%. Trading came to a complete stop in March 2014.
The demise of Mt. Gox happened was a two stage event. From 15May 2013 and beyond there were reports of delayed and halted dollar withdrawals from the exchange. After the US government seized over $5 million in the months of May and June, Mt. Gox assets dollar withdrawals came to a virtual standstil.
Costumers could still trade, but transferring Mt. Gox dollar accounts away from the exchange was near impossible. (McMillan, 2013)
The second stage started on 7 February 2014 when Mt. Gox halted bitcoin withdrawals from the site citing technical problems with withdrawals. (Clinch, 2014). Whereas it was previously possible to withdraw bitcoins from the
exchange and spend them at one of the vendors accapting bitcoin or exchanging it for fiat money on an alternative exchange, this had now become impossible. Heavy delays with dollar withdrawals were still ongoing and the price on the Mt. Gox exchange started to drop. On 24 february 2014 Mt. Gox halted all trading and stopped processing orders. It filed for bankruptcy protection in Japan on 28 february. (Sidel, 2014)
The bankruptcy of Mt. Gox wasn’t the first time a bitcoin exchange ceased operations. Moore and Christin (2013) analysed 40 different bitcoin exchanges from the start of their operations until January 2013. In that time period 45% of the exchanges closed, 17 of the 19 were relatively small with a trade volume of less than 1,000 bitcoins per day. BitcoinMarket and TradeHill, the two larger exchanges, closed after their security had been compromised.
The bitcoin market therefore couldn’t have been unaware of the risk involved with exchanges. However, there was no exchange that approached the size of the Mt. Gox exchange. This research will look at the effect of the closure of the Mt. Gox exchange on the bitcoin exchange price returns, the spread between exchanges as well as market its effect on market efficiency as described below.
Eugene Fama pioneered the efficient market hypothesis. Three types of market efficiency are defined: strong, semi-strong and weak. The efficient market hypothesis assumes there is no arbitrage possible based on the information available since the price should already reflect this information. The strong form requires the price of a security to reflect all information, even information that is not available to the general public. The semi-strong form efficiency is achieved when the price of a security reflects all publicly known information and weak form efficiency is achieved when the price reflects all previous trade data. (Fama, 1970)
Research question and hypotheses
Greco (2001) reasons that new currencies strongly rely on trust for their acceptance and survival. The user of the currency has to be convinced the currency will have value in the future and that he can use the currency to trade goods and services.
The demise of the Mt. Gox exchange is the failure of the largest bitcoin associated company yet. Does the bankruptcy of the second oldest and in January 2013 still by far the most popular exchange in the US dollar to bitcoin market impact the overall trust in the bitcoin network? To find out the effect of the demise of Mt. Gox the following research question is to be answered in this research:
Is the return on the US Dollar to Bitcoin exchange rate negatively affected by the demise of the Mt. Gox exchange?
To answer the research question the following hypotheses are tested. Given the fact that, in period 2, the dollar withdrawals were halted and/or severely delayed it is hypothesised that (1) the bitcoin exchange rate return is higher on Mt. Gox than on the Bitstamp and BTC-E exchange in period 2. (2) A significant spread between the Mt. Gox exchange and the Bitstamp and BTC-E exchange develops in period 2. The delays in the Mt. Gox dollar withdrawals function as market friction and investor expect to be compensated for the additional time it takes to pay out in dollars. Additionally the dollar account holders who fail at getting their dollar accounts withdrawn from Mt. Gox have no other option than to buy bitcoins on the Mt. Gox exchange and transfer them to another. This is expected further stimulate the Mt. Gox exchange rate and the spread between Mt. Gox and Bitstamp and BTC-E. Given the halt of the bitcoin withdrawals from Mt. Gox in period 3 the only way to possibly transfer funds away from the Mt. Gox exchange is through the delayed dollar withdrawals. By the beginning of period 3 it was clear Mt. Gox was in trouble, the investors only chance to get funds out of the Mt. Gox was to exchange them for dollars on the Mt. Gox
exchange, this lead to an excess of bitcoins on the Mt. Gox market and a decline in bitcoin exchange price. Therefore it is hypothesised that (3) the exchange return
rate on Mt. Gox is lower than the exchange rate return of Bitstamp and BTC-E during period 3. The excess supply of bitcoins will drive the exchange price on Mt. Gox down and increase the spread between Mt. Gox and the other two exchanges. Therefore it is expected (4) that the spread between Mt. Gox and Bitstamp and is significantly higher in period 3 than in period 1. As soon as the element of uncertainty leaves the market at the end of period 3 it is expected that (5) the spreads between Bitstamp and BTC-E become statistically insignificant.
Methodology
The following regressions will be performed to examine the effects of the Mt. Gox problems on the exchange price return and the exchange rate spread between Mt. Gox, Bitstamp and BTC-E. An ordinary least squares regression will be performed to estimate the linear effects of the variables on the return and spread. The first regression (1) is performed to determine the effect on the returns.
Rit = β0 + β1 Period 2 + β2 Period 3 + β4 Period 4 + β5 RS&P500,t + β6 Velocityt + β7 Viewst + εi (1) The second regression (2) is performed to determine the effect on the spread.
Spreadijt = β0 + β1 Period 2 + β2 Period 3 + β4 Period 4 + β5 RS&P500,t + β6 Velocityt + β7 Viewst + εi (2) Where Rit is the daily exchange return for exchange i on day t, the exchange
return is taken from the US Dollar to Bitcoin exchange rate of three different exchanges, Mt. Gox, Bitstamp and BTC-E. Spreadijt is the spread of the price
between exchange i and j, the spread is calculated by subtracting the exchange price on exchange i for day t from the exchange price of exchange j for day t.
Period 2, 3 and 4 are dummy variables for the corresponding time period, if all dummy variables are null the result for period 1 is shown, the dummy variables show the deviation of returns (1) and spread (2) from period 1.
Velocityt is the velocity of bitcoin; the frequency each bitcoin is used for
transactions on a particular day. Viewst are the number of times the Wikipedia
article about Bitcoin is viewed on a single day. RS&P500,t is the daily return on the
S&P500 index.
The Mt. Gox data will give insight in the way the market responds to the problems regarding the withdrawals. The Bitstamp and BTC-E exchange are used to determine the return and spread on exchanges that are not experiencing problems with withdrawals.
The control variables Velocityt, Wikipedia Viewst and RS&P500 are used to correct
for the effects of velocity of money; the demand for information about Bitcoin, and developments in the US market respectively.
Velocityt is the daily bitcoin velocity and measures how often one bitcoin is used
for a transaction per day. The higher the velocity, the more transactions are being completed using bitcoin. Velocity is calculated using the bitcoin value of transactions per day divided by the total amount of mined bitcoin up until that point. The velocity variable measures both changes in the bitcoin money supply and the changes in the bitcoin volume of transactions. The quantity theory of money suggest that if all else equal an increase in the velocity of money results in the increase in the price level. The inflation will result in a decrease in the value of bitcoin. It is expected that a higher velocity will result in a lower exchange rate and thus a lower return on the US dollar to bitcoin exchange rate. (Duck, 1993)
The velocity of bitcoin is computed using data from Blockchain.info. The total bitcoin value of all transactions on a certain day is divided by the amount of bitcoins in circulation on that day to calculate the velocity. The transaction value includes change returned to the sender and is therefore overestimated. The amount of bitcoins in circulation grows at a predetermined rate. The amount does not take into consideration the amount of bitcoins that have been lost; if the key to a bitcoin wallet is lost, the bitcoins are not retrievable. The actual number of bitcoins in circulation is expected to be lower than the amount of the total
bitcoins mined since the first bitcoins were mined in 2009. Both numerator en denominators seem to be overestimated; therefore it is assumed the measure accurately estimates the velocity of bitcoin.
Wikipedia viewst are the daily page views of the bitcoin Wikipedia page. The
variable can be used to estimate the demand for information about the currency. (Ciaian, Rajcaniova, & Kancs, 2014) A study by Kristoufek (2013) found that an increase in the public interest can lead to a higher bitcoin price. Kristoufek also uses the Google Trend data, which according to the study is a better predictor for the price movements. However, Google Trend data is only available on a week-by-week basis, whereas the dataset in this research is composed of daily data, therefore the Google Trend data is not used.
The S&P 500 is a weighted index of 500 companies on the traded U.S. equity market. The weight of each company on the index is derived from the market value of that company. The S&P 500 is a benchmark for the U.S. market and in large the U.S. economy. (Chiarella & Gao, 2004) The S&P 500 is used to capture the changes in the U.S. market, which might influence the price of bitcoin. The influence could be through substitution, a downturn in the U.S. economy would result in investors choosing a different investment opportunity, such as bitcoin; or complementary in that a downturn in the U.S. economy would result in investors moving money away from risky investments such as bitcoin to less volatile products.
Data
The data is collected for a 17-month time period, from 1 January 2013 to 1 June 2014, separated into four different periods. Summary statistics of the data are provided in table 1.
Table 1: Summary statistics exchange prices and spreads.
Mt. Gox exchange price Spread Mt. Gox - Bitstamp
Average SD Max Min Average SD Max Min
Period 1 $62.67 $48.68 $214.67 $13.31 $0.78 $4.82 $52.62 $-2.70 Period 2 $366.42 $354.11 $1,209.94 $70.26 $31.03 $41.19 $196.19 $-1.27 Period 3 $394.86 $195.66 $729.09 $119.34 $-239.71 $158.91 $26.50 $-443.45
Period 4 - - - -
Bitstamp exchange price Spread Mt. Gox - BTC-E
Average SD Max Min Average SD Max Min
Period 1 $61.89 $47.74 $214.86 $13.01 $3.17 $6.18 $14.14 $-0.69 Period 2 $335.38 $319.94 $1,132.29 $67.21 $38.01 $46.53 $97.79 $-4.20 Period 3 $634.57 $41.19 $702.59 $562.79 $-228.72 $154.29 $33.56 $-426.86 Period 4 $517.03 $77.39 $680.79 $392.23 - - - -
BTC-E exchange price Spread Bitstamp - BTC-E
Average SD Max Min Average SD Max Min
Period 1 $59.50 $45.35 $202.82 $13.11 $2.40 $3.15 $-4.76 $-4.76 Period 2 $328.40 $312.84 $1,053.57 $66.67 $6.98 $16.81 $-10.35 $-10.35 Period 3 $623.58 $48.51 $717.55 $546.20 $10.99 $13.44 $-14.96 $-14.96 Period 4 $509.86 $76.43 $669.88 $392.85 $7.17 $6.82 $-16.85 $-16.85
The first period is used as a benchmark for the rest of the data; the period starts at 1 January 2013 and lasts until 14 May 2013. The second period lasts from 15 May 2013 to 6 February 2014, during which there were significant delays and occasional halts in the withdrawal of dollar funds from Mt. Gox. The third period starts at 7 February 2014 when the withdrawal of both dollar and bitcoin funds from Mt. Gox are halted, the period lasts until 24 February 2014, the day on which all trading was permanently halted on Mt. Gox. The last period is the period after Mt. Gox trades are halted and lasts from 25 February 2014 to 1 June 2014.
During the data collection period the exchange price of bitcoin appreciated from $13 on all three exchanges on 1 January 2013 to $638 at the BTC-E and $646 at the Bitstamp exchange on 1 June 2014, an appreciation of 4730% on and 4870% respectively. By that time Mt. Gox had already stopped trading and had filed bankruptcy protection. The last day on which Mt. Gox processed transactions, 25 February 2014, the value had dropped to $123 per bitcoin, a fraction of the highest value of $1209 on 30 November 2013.
The returns fluctuated heavily per day; on the Mt. Gox exchange the maximum appreciation on one day was 77.81%, whereas the Bitstamp and BTC-E exchange saw a more humble maximum daily appreciation of 33.75% and 25.37%
respectively. Along with upward price movement, there were a few significant downward movements as well; the highest daily depreciation was 46.16%, 39.12% and 44.22% on the Mt. Gox, Bitstamp and BTC-E exchange respectively.
The returns on the exchange prices are strongly correlated as is expected, because they trade the same product. The law of one price states that the
homogenous nature of bitcoin, one bitcoin is no different from the other; bitcoins across different exchanges should be similarly priced. The correlation of the exchange rates during the four periods is shown in table 2 in appendix 1. The correlation in period 3, however, is considerably lower than the correlations in the other periods. The correlation between Mt. Gox and BTC-E is 0.887, whereas the correlation between Mt. Gox and Bitstamp is 0.914. The correlation between the exchanges without withdrawal problems is 0.968.
The exchange price development over time in figure 1 shows a similar story. The exchange price is relatively similar during the first period, as shown in table 1 the average spread between $0.78 and $3.17 depending on the exchange. The Bitcoin exchange price seems to be higher on Mt. Gox during the second half of period 2 and lower during period 3 compared to the other two exchanges. Table 1 shows that the average spread during period 2 between Mt. Gox and Bitstamp is $31, meaning the price on Mt. Gox is on average $31 higher than on Bitstamp. The spread between Mt. Gox and BTC-E for that same period is $38, this is in line
with our expectations as voiced in hypothesis (2). The data also seems to
support hypothesis (4). The spread between Mt. Gox and Bitstamp is on average $ -240 and the spread between Mt. Gox and BTC-E $229, during the same period the spread between Bitstamp and BTC-E was $11, during period 4 this spread decreases to an average of $7, higher than expected in hypothesis (5).
Figure 1: Historical price development on three bitcoin exchanges. First grey band is period 1, first white band is period 2, second grey band is period 3 and second white band is period 4.
The returns on the exchange price of bitcoin are in strong contrast with the returns of the S&P 500. Whereas the daily average change on the bitcoin exchanges hovers around 0.9%, the mean daily return on the S&P 500 is only 0.09%. The risk, as measured by the standard deviation, is significantly lower as well, the standard deviation for the daily returns of the S&P 500 are 0.7%
significantly lower than the 8.8%, 6.4% and 6.1% standard deviation of the returns on the bitcoin exchanges. The S&P 500 is relatively stable with a highest change in value in both directions of around 2.5%.
The velocity of bitcoin has a mean of 0.09 meaning on average 9% of all bitcoins are spent in a day. This measure is significantly higher than the velocity of
traditional currencies, the yearly velocity of bitcoin given the mean of 0.09 would come near 33, whereas the federal reserve bank of St. Louis reported a velocity of 1.5 for the US dollar in 2013. The reason for difference is in part difference in the way the velocity is computed. The velocity is usually calculated using the valued added transactions as opposed to all transactions. Since every bitcoin transaction is included in the calculation of the bitcoin velocity, the velocity measure is heavily overestimated. Assuming that the percentage of added value transactions per total transactions stays equal the measure can still be used as a control variable since we’re interested in the effect of the change in velocity on the exchange price, not the effect of the absolute value of the velocity on the exchange price.
On Wikipedia the page about Bitcoin was viewed an average of 21,000 times per day, with a standard deviation of 20,000 views. The amount of views fluctuates heavily on a day-by-day basis, with a maximum of 142,760 views on 11 April 2013 and a minimum number of 2671 views on 1 January 2013.
Results:
The regression results for regression (1) are summarised in table 3, the results for regression (2) are summarised in table 4. The regressions show signs of heteroskedasticity, both the Breusch-Pagan / Cook-Weisberg test for
hetereoskedasticity and the White’s test for homoskedasticity reject the null hypothesis of homoscedasticity. The results of the Breusch-Pagan / Cook- Weisberg test and the White test are found in table 5 in appendix 1.
In case of heteroskedasticity the ordinary least-squares estimators are still unbiased, however the standard errors are underestimated. As a result the risk of type I errors increase, which means a risk of finding statistical relationships that are actually non-existent. (Osborne & Waters, 2002) To compensate for the heteroskedasticity the heteroskedasticity proof standard errors are used while
estimating the regression. The results using these standard errors can be found under the ‘robust‘ specification.
The regression (1) is used to estimate the effect of the different periods on the exchange rate of return. A higher return on the exchange rate was expected for Mt. Gox in period 2; however there is no statistical evidence that period 2 saw significantly higher returns than in period 1. Therefore hypothesis (1) must be rejected. The regression shows no significantly higher returns for Bitstamp and BTC-E either, this is as expected since the two exchanges were unaffected by withdrawal problems during this period.
Regression (2) is performed to check the effect of time on the spread. The results show that Mt. Gox saw the spread with other exchanges rise from trading at a discount in period 1, to trading at a significant premium in period 2. This is in line with the expectations and in line with the semi-strong form of the efficient market hypothesis (Fama, 1970). The information about the delays in dollar withdrawals is public and seems to have caused a difference in pricing between two homogenous goods. The price differential can only be explained by market failure, which in this case would be caused by transaction delays. Furthermore the withdrawal problems prevent the possibility of arbitrage. Dollars deposited in Mt. Gox accounts have a lower value than regular dollars since they can only be used to buy bitcoin, the market is aware of this and values dollars on Mt. Gox lower than regular dollars, in return the exchange rate of bitcoin on the Mt. Gox exchange is higher.
Evidence is found for a lower exchange rate return in period 3 for Mt. Gox. The bitcoin and dollar withdrawal problems cause a drop in the expected return of 14.8%. The halt of the bitcoin withdrawals changed the relative value of bitcoins and dollars on Mt. Gox. Whereas in period 2 dollars were worth less on Mt. Gox than on other exchanges, because of the delays and partial halt in dollar
withdrawals, the complete halt in bitcoin withdrawals has made dollar
withdrawals the only viable, if unlikely, exit option. The relative value of dollar rise and the exchange rate for bitcoin drops on the Mt. Gox exchange. The
Bitstamp and BTC-E exchanges see no significant reduction in their exchange rate returns. No evidence is found that the problems of the Mt. Gox exchange influence the exchange prices at Bitstamp or BTC-E, in that regard it can be said that the Mt. Gox crisis is a contained crisis.
Further evidence for the containment is found when looking at the spread in period 3. The spread between Mt. Gox and Bitstamp and BTC-E has risen
significantly. Whereas in period 2 the exchange rate on Mt. Gox was $16 and $18 higher than the Bitstamp and BTC-E exchange price respectively. The period 3 spread has changed considerably, compared to Bitstamp and BTC-E, bitcoin on Mt. Gox is worth between $254 and $250 less. The spread is clear evidence of the dumping of bitcoins for dollars, which could potentially be withdrawn from the Mt. Gox accounts. The spread between Bitstamp and BTC-E is $4.50, insignificant compared to the Mt. Gox spread, but still large for a homogenous good. The spread is significantly higher than the spread in period 1, which can be seen as an indicator that the Mt. Gox crash in period 3 is having some effect on the efficiency of the exchange market.
Period 4 shows no significant deviations in return for Bitstamp or BTC-E either. The effect of the Mt. Gox crash seems to have an insignificant effect on the
exchange price return. The spread between Bitstamp and BTC-E, which had risen in period 3 is back to a spread of $1.14, given the average exchange price of $519 in period 4, the spread is a mere 0.2%. Given the minimum transaction costs of 0.2% on Bitstamp this spread can be attributed to transaction costs. (Bitstamp)
The regressions provided evidence that the demise of the Mt. Gox exchange has not influenced the returns on the exchange rate in any significant way. Although the spread between Bitstamp and BTC-E grew larger at the time of the Mt. Gox crisis, the spread returned to a lower level after the Mt. Gox exchanged exited the market. No evidence has been found of the crisis affecting the value of bitcoin in a negative way.
Table 3: Regression (1) of exchange rate returns over periods 1 - 4
*** = 1% significance level, ** = 5% significance level * = 10% significance level Coefficient and (Standard Deviation)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)
Mt. Gox Mt. Gox Mt. Gox Mt. Gox Mt. Gox Bitstamp Bitstamp Bitstamp Bitstamp Bitstamp BTCE-E BTCE-E BTCE-E BTC-E BTCE-E
Robust Robust Robust
Period 2 -.011 -.011 .007 .014 .014 -.117* -.010 .004 .0123 .0123 -.0108* -.0100 .0064 .0130 .0130 (.010) (.012) (.012) (.016) (.007) (.009) (.010) (.010) (.014) (.006) (.008) (.010) (.010) (.013) Period 3 -.073*** -.186*** -.161 -.148*** -.148*** -.037** -.058*** -.038* -.023 -.023 -.0343** -.0577*** -.0349 -.0224 -.0223 (.024) (.026) (.026) (.041) (.0159) (.022) (.023) (.023) (.020) (.015) (.021) (.022) (.023) (.023) Period 4 - - - -.018** -.019* -.0013 .006 .006 -.0176** -.0212** -.00124 .0050 .0050 (.008) (.011) (.013) (.013) (.014) (.008) (.011) (.013) (.013) (.013) S&P500 -.771 -.833 -.916 -.916 -.302 -.340 -.426 -.426 -.3427 -.3864 -.4546 -.4546 (.630) (.623) (.620) (0.683) (.513) (.509) (.503) (.586) (.501) (.497) (.493) (.562) Velocity .230*** .297*** .297*** .188*** .269*** .269 .2118*** .2797*** .2797*** (.087) (.090) (.089) (.075) (.077) (.079) (.073) (.075) (.079) Views -0.005** -0.005 -.006*** -.006*** -.0050*** -.0050 (.0021) (0.0042) (.0017) (.0037) (.002) (.004) Constant .020*** .023*** -.013 -.013 -.013 .020*** .022*** -.008 -.008 -.008 .0195*** .0224*** -.0113 -.0113 -.0112 (.008) (.016) (.016) (.017) (.005) (.007) (.014) (.014) (.016) (.005) (.007) (.013) (.013) (.015) N 420 288 288 286 286 517 355 355 353 353 517 355 355 353 353 R2 0.0268 0.1803 0.2002 0.2189 0.2189 0.0157 0.0258 0.0433 0.0776 0.0776 0.0154 0.0284 0.0514 0.0764 0.0764 R2 adjusted 0.0221 0.1716 0.1889 0.2049 0.0100 0.0147 0.0296 0.0616 0.0097 0.0173 0.0378 0.0604
Table 4: Regression (2) of exchange price spreads over periods 1 - 4
*** = 1% significance level, ** = 5% significance level * = 10% significance level Coefficient and (Standard Deviation)
(1) Mt. Gox - Bitstamp (2) (3) Mt. Gox - BTC-E (4) (5) Bitstamp - BTC-E (6)
Robust Robust Robust
Period 2 20.39*** 20.39*** 28.46*** 28.46*** 7.91*** 7.91*** (6.50) (3.11) (6.84) (3.69) (1.72) (2.04) Period 3 -251.62*** -251.62*** -237.76*** -237.76*** 13.76*** 13.76*** (14.70) (47.60) (14.90) (47.09) (3.76) (4.58) Period 4 10.31*** 10.31*** (2.11) (2.04) S&P500 37.59 37.59 0.46 0.46 -20.84 -20.84 (349.67) (380.37) (354.53) (381.51) (82.48) (56.38) Velocity -88.87** -88.87** -30.64 -30.64 54.74*** 54.74*** (50.60) (34.93) (51.30) (33.83) (12.61) (19.17) Views 8.07*** 8.07*** 9.78*** 9.78*** 1.41*** 1.41*** (1.16) (1.30) (1.18) (1.58) (.27) (0.5) Constant -2.23 -2.23 -12.56 -12.56** -9.17*** -9.17*** (8.89) (4.57) (9.01) (4.92) (2.23) (3.3) N 286 286 286 286 353 353 R2 0.6233 0.6233 0.6246 0.6246 0.1845 0.1845 R2 adjusted 0.6165 0.6179 0.1703
Conclusion
Bitcoin is a newcomer in the currency market that deserves our attention. The decentralised framework, the availability and traceability of transaction data and the controlled money supply make bitcoin an interesting research subject.
The bitcoin exchange market is a market with high daily returns and high daily risk. Compared to the S&P500 both the volatility and returns are significantly higher.
The exchange price on the Mt. Gox exchange was severely affected during the withdrawal problems. Dollar withdrawal problems caused the spread to grow with $20 to $28 between the Mt. Gox exchange and the Bitstamp and BTC-E exchange respectively. Bitcoin withdrawal problems pushed the spread to more than $250.
The eventual bankruptcy led to significantly lower returns on the bitcoin
exchange rate, however only for the Mt. Gox exchange. The Bitstamp and BTC-E exchange show no signs of being affected by the withdrawal problems at Mt. Gox and its eventual demise. No significant evidence is found indicating the bitcoin exchange rate was affected by the Mt. Gox crisis. Some evidence is found that the Mt. Gox crisis affected the market efficiency, however the effect was only
Literature
Barber, S., Boyen, X., Shi, E., & Uzun, E. (2012). Bitter to better—how to make bitcoin a better currency. Financial Cryptography and Data …, 399–414. Retrieved from http://link.springer.com/chapter/10.1007/978-3-642-32946-3_29
Bitstamp. Retrieved June 17, 2014, from
https://www.bitstamp.net/fee_schedule/
Blockchain.info Retrieved June 17, 2014, from Blockchain.info/charts
Chiarella, C., & Gao, S. (2004). The value of the S&P 500—A macro view of the stock market adjustment process. Global Finance Journal, 15(2), 171–196. doi:10.1016/j.gfj.2004.03.002
Christin, N. (2013). Traveling the Silk Road: A measurement analysis of a large anonymous online marketplace. Proceedings of the 22nd International
Conference on …, 1–26. Retrieved from
http://dl.acm.org/citation.cfm?id=2488408
Clinch, M (2014) Bitcoin plummets 20% after major exchange halts withdrawals
CNBC. Retrieved from:http://www.cnbc.com/id/101398515
Ciaian, P., Rajcaniova, M., & Kancs, A. (2014). The Economics of BitCoin Price Formation. arXiv Preprint arXiv:1405.4498. Retrieved from
http://arxiv.org/abs/1405.4498
Demirgüç-Kunt, A., & Levine, R. (1996). Stock market development and financial intermediaries: stylized facts. The World Bank Economic Review. Retrieved from http://wber.oxfordjournals.org/content/10/2/291.short
Duck, N. (1993). Some international evidence on the quantity theory of money.
Journal of Money, Credit and Banking, 25(1), 1–12. Retrieved from
http://www.jstor.org/stable/2077816
European Central Bank. (2012). Virtual Currency Schemes (pp. 1 – 55). Retrieved from
http://www.ecb.europa.eu/pub/pdf/other/virtualcurrencyschemes201210 en.pdf
Fama, E. (1970). Efficient capital markets: A review of theory and empirical work*. The Journal of Finance. Retrieved from
http://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1970.tb00518.x/full
Gomez-Gonzalez, J., & Parra-Polania, J. (2014). Bitcoin: something seems to be “fundamentally” wrong. Borradores de Economía, (819), 1–9. Retrieved from http://www.banrep.gov.co/sites/default/files/publicaciones/archivos/be_ 819.pdf
Goodhart, C. (1995). Central bank and the financial system. arXiv preprint
arXiv:1402.4778 (pp. 1–26). Retrieved from
http://arxiv.org/abs/1402.4778
Greco, T. (2001). Money: understanding and creating alternatives to legal tender.
The Journal of Political Economy (Vol. 72, pp. 584–596). Retrieved from
http://digamo.free.fr/balassa64.pdf
Grinberg, R. (2012). Bitcoin: an innovative alternative digital currency. Retrieved from
http://heinonlinebackup.com/hol-cgi-bin/get_pdf.cgi?handle=hein.journals/hascietlj4§ion=6
Hanley, B. (2013). The False Premises and Promises of Bitcoin. arXiv Preprint
arXiv:1312.2048, 1–32. Retrieved from http://arxiv.org/abs/1312.2048
Historyofbitcoin.org. Retrieved June 17, 2014, from Historyofbitcoin.org Kristoufek, L. (2013). BitCoin meets Google Trends and Wikipedia: quantifying
the relationship between phenomena of the Internet era. Scientific Reports,
3, 3415. doi:10.1038/srep03415
Kristoufek, L. (2014). What are the main drivers of the Bitcoin price? Evidence from wavelet coherence analysis. arXiv Preprint arXiv:1406.0268, 1–19. Retrieved from http://arxiv.org/abs/1406.0268
MacDonell, A. (2014). Popping the Bitcoin Bubble: An application of log-‐periodic power law modeling to digital currency, (December 2013). Retrieved from http://economics.nd.edu/assets/134206/mac_donell_popping_the_biticoin_ bubble_an_application_of_log_periodic_power_law_modeling_to_digital_curr ency.pdf
Mcmillan, R (2013) The rise and fall of the world's largest bitcoin exchange.
Wired. Retrieved from http://www.wired.com/2013/11/mtgox/
Moore, T., & Christin, N. (2013). Beware the middleman: Empirical analysis of Bitcoin-exchange risk. Financial Cryptography and Data Security, (June 2011), 25–33. Retrieved from
http://link.springer.com/chapter/10.1007/978-3-642-39884-1_3 Moser, M., Bohme, R., & Breuker, D. (2013). An inquiry into money laundering
tools in the Bitcoin ecosystem. eCrime Researchers Summit ( …. Retrieved from http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6805780
Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. Consulted, 1–9. Retrieved from http://s.kwma.kr/pdf/Bitcoin/bitcoin.pdf
Osborne, J., & Waters, E. (2002). Four assumptions of multiple regression that researchers should always test. Practical Assessment, Research, and
Evaluation, 8(2), 1–5. Retrieved from
http://www-psychology.concordia.ca/fac/kline/495/osborne.pdf
Sidel, R (2014) Almost half a billion worth of bitcoins vanish. The Wall Street
Journal. Retrieved from
http://online.wsj.com/news/article_email/SB100014240527023038013 04579410010379087576
Wallace, B. (2011). The rise and fall of Bitcoin. Wired Magazine.[Online].
Available, 1–9. Retrieved from
http://fromm.robertkeahey.com/wp-content/uploads/2013/07/Session-7-The-Rise-and-Fall-of-Bitcoin.pdf Yermack, D. (2014). Is bitcoin a real currency? An economic appraisal.
Appendix 1
Table 2: Correlations of Bitcoin exchange prices, in periods 1, 2, 3 and 4 on Mt. Gox, Bitstamp and BTC-E Mt. Gox Bitstamp BTC-E
1 Mt. Gox 1.000 - - Bitstamp 0.995 1.000 - BTC-E 0.994 0.999 1.000 2 Mt. Gox 1.000 - - Bitstamp 0.998 1.000 - BTC-E 0.998 0.999 1.000 3 Mt. Gox 1.000 - - Bitstamp 0.914 1.000 - BTC-E 0.887 0.968 1.000 4 Mt. Gox - - - Bitstamp - 1.000 - BTC-E - 0.996 -
Table 5: Breusch-Pagan / Cook-Weisberg and White’s tests for heteroskedasticity on regressions (1) and (2)
Regression (1)
Model (4) (9) (14)
Distribution Chi2 Chi2 Chi2
BP / CW 35.18 *** 34.99*** 9.98***
White 107.54*** 182.68*** 173.88***
Regression (2)
Model (1) (3) (5)
Distribution Chi2 Chi2 Chi2
BP / CW 659.60*** 485.88*** 508.67***
White 198.73*** 204.33*** 92.97***