The purpose of this article is to provide some insights on the true nature of bitcoin and to study empirically its performance by using robust models, widely used in the academic literature. Previous studies assess performance with simple measures such as the Sharpe ratio. Such measures are insufficient because they do not take into account the bitcoin’s specificities, such as the possibilities to diversify risk.
The authors use quantitative methodologies to assess the performance of financial assets. Performance is defined as a risk-adjusted return. The authors use regression analysis and measure bitcoin’s performance as the constant term (α) of the projection of its returns on the returns of relevant factors of risk.
Bitcoin has low correlation with the market index and with factor-mimicking portfolios, which indicates opportunities to diversify risk. The performance of bitcoin (α) is positive and significant; this result is robust across period and world region specifications.
The true nature of bitcoin is subject of debate and needs further research. Furthermore, other factors should be considered in analysing the bitcoin’s performance, such as those related to investors’ behaviour or political risk.
The empirical results obtained in this paper may be used by professional portfolio managers to diversify risk and to enhance their portfolio’s performance.
This paper adds to the literature by arguing that bitcoin has the nature of common stock, and therefore, its performance has to be assessed with models that are relevant for this type of securities. This paper is the first using performance models that adjust returns for relevant sources of risk.
CitationDownload as .RIS
Emerald Publishing Limited
Copyright © 2019, Emerald Publishing Limited
Bitcoin is a cryptocurrency, and therefore a part of the so-called “electronic currency” family (Nakamoto, 2008). On 04 January 2017, the bitcoin price was equal to US$1,126, thus reaching the threshold of US$1,000 for the second time in its history. Incredibly unexpected, and even more so, the price raised to more than US$10,000 one year after reaching US$16,808 on 13 December 2017.
Electronic currencies, which are recorded on a computer or other electronic devices, may take two forms: digital cash and cryptographic currency. The former is a simple digital version of physical money. The latter, which is studied in this article, uses cryptographic principles to ensure security. This type of currency is developed in a decentralised system that uses the blockchain technology.
Whilst bitcoin has common characteristics with currencies (bitcoin is indeed an exchange medium), and with gold (the monetary creation of bitcoin looks like that concerning gold), we hereafter consider bitcoin as a financial contract, and more specifically common stock. Bitcoin indeed has specific economic and legal profiles, risk-return characteristics and its liquidity bring it close to the common stock family (Glaser et al., 2014; Baur et al., 2018).
The objective of this article is to study empirically the bitcoins’ performance. The justification of studying the performance of bitcoin relies on the fact that it is a relatively young, risky financial innovation, not yet well assimilated by the market. Bitcoin is affected by informational asymmetries between investors, mainly because many of them do not understand the technology behind it. Consequently, the market value of this financial contract may differ from its true value, that is, the value on a market without informational asymmetries, thus providing opportunities to earn positive risk-adjusted returns. In line with this view, the objective of some mutual funds is to track bitcoins. This suggests the possibility of earning superior risk-adjusted returns by trading this financial contract.
In most existing articles, the performance of bitcoin is assessed by using simple measures, such as the Sharpe ratio (Brière et al., 2015; Burniske and White, 2017). The problem with this measure is that it does not take into account the huge potential for portfolio diversification with bitcoins. Being highly uncorrelated with most existing assets, most of the bitcoin variability can be diversified away, which strengthens its performance. Based on this important aspect, and on the premise that bitcoin has the nature of a financial contract, our paper is the first, to our knowledge, that attempts to measure its performance by using models such as the capital asset pricing model (CAPM), which uses the market portfolio as a benchmark, and also the three-factor Fama and French (1992) model, hereafter FF, and also other extensions of these models by considering other factors related to bitcoin.
To take into account the international dimension of bitcoin, we use the global market portfolio as a benchmark and the global versions of the FF factors (Fama and French, 2012). To test bitcoin performance, we compute the α estimated with these models for the period between 22 September 2010 and 30 December 2016 for different regions: World, Europe, Asia-Pacific and China. In addition, we take into consideration the market sentiment variable in our models to check the robustness of our results and to take into account behavioural aspects that are plausibly related to bitcoin.
The rest of the paper is organised as follow. Section 2 presents bitcoin characteristics. Section 3 describes the models used to assess bitcoin performance and the data sets used in our study. Section 4 discusses the results. Section 5 proposes some robustness tests related to the violations of the regression model assumptions, mainly the normality of errors, and also to some behavioural aspects. Section 6 presents the conclusion.
2. About the nature of bitcoin
The nature of bitcoin is subject of debate. Academics and professionals suggest various definitions and do not always agree on whether bitcoin is a currency, a commodity, a “safe” investment such as gold, a debt contract or common stock. In this section, we present bitcoin characteristics, and we attempt to show that bitcoin has also the characteristics of an asset, more precisely common stock (Glaser et al., 2014; Baur et al., 2018).
2.1 What is bitcoin?
Bitcoin is basically a payment system that is independent from any government and operates without a third party (such as a Central Bank). Any participant in this system may check the behaviour of other participants to ensure the reliability of transactions and system stability. In other words, the well-functioning responsibility is not assumed by a third party. Moreover, all participants have the possibility to know the transactions made by everyone. However, the identity of participants remains pseudo-anonymous, in the sense that the identity is hidden by a pseudonym, the bitcoin “address”, which is a number sequence.
There are three ways to obtain bitcoins: exchanging money, selling goods and services or mining. The first two ways, which are exchanging money and selling goods or services through e-businesses that accept bitcoin units, make bitcoin behave like a fiat currency. However, a fundamental aspect that makes bitcoin behave like a financial asset is that one may create bitcoins through the mining process, whose underlying technology is named “Blockchain”. This is a secured and distributed database that contains the history of transactions. The technology may be considered as a ledger that stores all exchanges realised on the network (Nakamoto, 2008; Tschorsch and Scheuermann, 2016). This ledger is composed of blocks linked to each other. Each block contains a transaction list of some exchanges. Special users, named miners, create a block locally by choosing different pending transactions. Then, each new block is drawn by a mathematical process, comparable to Sudoku. When a miner finds the grid solution, he/she wins a predetermined number of bitcoins, and other participants must start again the competition with another grid. Grid difficulty is adjusted so that miners may find the grid solution with an interval of 10 min on average between each discovery (Antonopoulos, 2015).
Miners are compensated for their work of integrating transactions to the blockchain and making the system reliable. They obtain a number of bitcoins when they succeed to add a block to the blockchain. This compensation is predefined in advance and the number of released bitcoins decreases with time because the maximum number of bitcoins is capped at 21 million.
2.2 Is bitcoin a currency?
From an economic perspective, the question whether bitcoin has common characteristics with currency is subject to debate. According to its basic properties, a currency should be a convenient medium of exchange, a stable unit of account and a durable store of value (Grant, 2014). Bitcoin is a medium of exchange, in the sense that a high number of businesses, such as Dell, Microsoft or PayPal, are willing to accept bitcoins (Figuet, 2016).
The popularity of using bitcoins is based on the user’s anonymity and the system transparency (because it uses the blockchain technology). However, participants could be reluctant to participate in this new system because bitcoin has no legal basis; companies make the choice to use bitcoin or not; the fixed costs of adopting this technology are high (sophisticated technical knowledge is needed); and there are network externalities effects (if few businesses accept bitcoin, few consumers may accept them, which in turn implies that few companies decide to accept them). Resolving this “vicious circle” is difficult because bitcoin is not regulated by an institution, and there is no possibility to make loans on the market (Kancs et al., 2015). Empirical studies confirm the controversial property of “medium of exchange” based on the fact that users do not entirely turn to bitcoin for this property (Baur et al., 2018).
Bitcoin may be considered to some extent as a “unit of account”. However, merchants do not display prices in bitcoin for two reasons. First, its supply is inelastic (a bitcoin price of a given product needs many digits after the comma). Second, volatility does not ensure price stability. Because of high volatility, merchants are forced to change frequently the price of their products. So, prices are usually displayed in US dollars and are then converted in bitcoins at the time when the transaction takes place (Figuet, 2016).
Finally, whilst bitcoin has features that make it behave like a “store of value”, the possible cybersecurity risks reduce trust in this currency. The trade-off between inflation and deflationary pressure, as well as the unstable purchasing power make it difficult to consider bitcoin as a store of value.
To conclude, bitcoin is different from traditional money because it does not fully respect the fiat currency properties, and in particular, there is no issuer responsible for it. In fact, bitcoin is governed by a protocol run by a network of computers that are distributed around the world; government monetary policies have no direct impact on it.
In economics, there are different perspectives related to a currency. Another point of view is to see currency through institutionalist economics as a “unit of account”. Based on this, bitcoin could be considered as a “currency” in the sense of money being a “social institution” (Lakosmki-Laguerre and Desmedt, 2015).
2.3 Is bitcoin “digital” gold or cash?
Some researchers consider the possibility that bitcoin acts like a commodity, more precisely as gold, for some reasons: supply is limited, meaning a weak inflation rate; monetary creation is based on the “mining” process; there is no control by any government; and bitcoin acts as a medium of exchange (Dyhrberg, 2016a, 2016b). Precious commodities such as gold are relatively safe, and bitcoin may often play a role of safe investment (Bouri et al., 2016). For example, after the Cyprus crisis in 2012-2013, some depositors exchanged euros for bitcoins because of the bankruptcy of banks and also because of increased taxes on deposits. The second peak at US$1,000 was linked to the international context in both developed and emerging countries. Chinese depositors turned toward bitcoin because of the Yuan drop and Chinese restrictions on capital outflows (this trend was accelerated by Donald Trump’s election in the USA). Likewise, unstable monetary policies in emerging countries (inflation in Venezuela, demonetisation in India, liquidity crisis in Zimbabwe) encouraged local depositors to turn to bitcoin. Empirically, bitcoin appears to be a weak safe investment reserved to some special cases but not during financial crises (Bouri et al., 2016; Baur et al., 2018).
While bitcoin may act as a safe investment in some cases, we can observe major differences with precious commodities such as gold. First, bitcoin is capped at 21 million and the release of new bitcoins is divided by two approximately every four years until the maximum number of bitcoins is reached. However, the supply for gold is not known. Moreover, even if the price of bitcoin increases, its capitalisation (US$117,353,600,811 in November 2017) is much lower than for gold (US$7,747,981,667,403 in November 2017). Second, the price of bitcoin is independent from that of gold. The factors affecting the value of these two assets are different. Finally, gold (or any commodity) has physical shape, whilst bitcoin possesses a virtual one.
An alternative complementary category is proposed by Selgin, a category named “Synthetic commodity money” that is both commodity money and fiat money (Selgin, 2015). In our case, bitcoin could be compared to gold for the former and to dollar for the later (Baur et al., 2018). Dyhrberg (2016a, 2016b) tests empirically the relationship between bitcoin and gold/dollar. Bitcoin appears to be positioned between gold (store of value) and dollar (medium of exchange).
Finally, bitcoin could also be assimilated to cash or to a cash equivalent. But a cash equivalent implies that the asset must be highly liquid and convertible into a known amount of cash. Bitcoin is convertible, but it is not enough liquid to be considered as cash equivalent. Bitcoin cannot be deposited into banks and cannot be withdrawn using ATM (Raiborn and Sivitanides, 2015).
2.4 Bitcoin as common stock
While bitcoin does not fully conform to the criteria of medium of exchange, store of value, or unit of account, the academic literature considers more and more the alternative view that bitcoin acts as an asset (Yermack, 2015). Amongst various assets, the similarity between bitcoin and common stock, which is the argument that we put forward in this paper, is justified by its high risk–high reward profile. Owning bitcoins implies owning a portion of:
the intangible asset represented by a specific technology, blockchain; and
the human capital represented by experts running code and using mathematical procedures to enhance the credibility of the system.
Bitcoin may therefore be considered as an investment that generates benefits in the same way as common stock.
Some empirical studies implicitly consider bitcoins as a financial contract because they test diversification possibilities with bitcoin or assess its performance through performance measures used for financial contracts, such as the Sharpe ratio (Bouri et al., 2016; Burniske and White, 2017). Glaser et al. (2014) show that users are not mainly interested in bitcoin as an alternative transaction system for paying goods or services, but rather as an alternative investment vehicle for transaction purposes, mostly for speculation. Two results contribute to this view. First, the interest of users influences the bitcoin’s volume traded at the exchange but has no effect on the volume within the bitcoin system. Second, bitcoin returns react on events related to the bitcoin’s security and infrastructure. While bitcoin seems to be mainly affected by positive events such as new exchange launches or legal successes, it is also affected by negative events such as system bugs, thefts, hacks or exchange breakdown.
In 2015, two mutual funds have been created to track bitcoins for investors: Bitcoin Investment Trust and ARK Investment Management. Bitcoin Investment Trust (BIT) works as an exchange trade fund (ETF) and tracks bitcoin, whilst ARK Investment Management creates two innovation funds which integrate the first one (BIT). The similarity between the bitcoin and financial contracts is also supported by the US Internal Revenue Service (IRS) in 2014. IRS considers bitcoin as property, and its holders are considered as market investors. After the DAO case, in 2017, the SEC announced that tokens linked to this project should have been considered as securities (common stocks) because the objective of these tokens is to generate earnings for investors taking risks to support the project.
3. Performance models
Existing academic and professional studies assess bitcoin performance mostly with simple measures such as the Sharpe ratio, which adjust returns for total (specific and diversifiable) risk. The volatility of bitcoin being very strong, this measure potentially underestimates performance. The low correlation of bitcoin with other existing assets and currencies implies a huge potential for diversification, which needs to be accounted for in performance measures. Furthermore, the international dimension of bitcoin is another important aspect that needs to be integrated into a performance analysis.
We estimate performance relative to the traditional CAPM and international benchmarks to take into consideration diversification effects. In addition, we estimate performance relative to the Fama and French (1992) three-factor model, which is widely used to assess the performance of financial contracts, such as the performance of actively managed portfolios. The use of common-stock-based indices to assess performance is justified by the similarities of bitcoin with common stock, as argued previously. We also use the return on gold and on bonds as additional factors in our performance models, as it is common in the existing literature. The specifications of the three models are as follows:
The risk-adjusted return generated by bitcoin is measured by the constant coefficient (α) from estimating the models specified by equations (1), (2) and (3). We consider global benchmarks because bitcoin is used across the world and is not linked to a special country. In their paper published in 2012, Fama and French apply their three- and five-factor models at an international level. They construct global portfolios using global size and B/M breakpoints for four regions to allocate the stocks of these regions to size and B/M portfolios (Fama and French, 2012).
The data for benchmarks’ returns are taken from the Kenneth R. French’s website for both global, European and Asia-Pacific analyses. The world region is composed of Australia, Austria, Belgium, Canada, Switzerland, Germany, Denmark, Spain, Finland, France, Great Britain, Greece, Hong Kong, Ireland, Italy, Japan, The Netherlands, Norway, New Zeeland, Portugal, Sweden, Singapore and the Unites States. The European region contains Austria, Belgium, Switzerland, Germany, Denmark, Spain, Finland, France, the Great Britain, Greece, Ireland, Italy, The Netherlands, Norway, Portugal and Sweden. The Asian-Pacific region refers to Australia, Hong Kong, New Zealand and Singapore. Data for gold and bond indices benchmarks (gold bullion USD/troy ounce rate, and Pimco Invest Grade Corporate Bond Exchange-Traded Fund, respectively) come from the Datastream database. Finally, bitcoin prices come from the blockchain info website: https://blockchain.info.
We estimate the above-specified models over the period from 22 September 2010 to 30 December 2016. The starting date is relatively recent because bitcoin is a recent innovation. We use daily returns to have a sufficient number of observations for our regressions, and thus obtain 1,638 daily observations over the specified period.
We also replicate our analyses for the Chinese market, where more than 90 per cent of the transactions on bitcoin occur. The data for this market are extracted from Datastream. We construct the size and B/M factor-mimicking portfolios on the Chinese market by following the same procedure as the one used by Fama and French for constructing their factor-mimicking portfolios on the US market.
4.1 Univariate results
Figure 1 presents the market price of the bitcoin over the analysis period, whilst Table I presents descriptive statistics for bitcoin, common stock indices of selected countries, MSCI indexes (for the world, Europe and Asia-Pacific regions), commodities indexes (oil, gold and commodity index), the bond index (Pimco) and currencies (dollar index, Yen, euro and Yuan). The bitcoin’s risk-return profile is, not surprisingly, very atypical, with a mean annualised return of 568.82 per cent for an SD of 111.69 per cent and a high spread between minimum and maximum returns. These results are similar to those found in the literature (Brière et al., 2015; Bouri et al., 2016; Baur et al., 2018). For comparison, the S&P 500 exhibits an annual return of 19.74 per cent for an SD of 17.73 per cent. The dollar index has a mean return of 25.15 per cent for an SD of 30.43 per cent and gold has a mean return of –2.14 per cent for an SD of 20.50 per cent.
We also note that bitcoin returns exhibit a positive skewness of 0.68, meaning they are skewed to the right. Almost all the other variables exhibit a negative skewness, except oil, and two currencies, Yen and Yuan. Note that existing empirical studies on bitcoin obtain positive or negative skewness coefficients. Furthermore, bitcoin has a heavy-tailed distribution with a kurtosis of 15.98 which indicates a higher number of tail events in bitcoin returns (Baur et al., 2018). This result is similar to others found in the literature. The empirical distribution of the returns is presented in Figures 6 and 7. Normality tests such as Shapiro–Wilk, presented in Table X, reject the null hypothesis of normal distribution.
Table II presents the Pearson correlation coefficients. Not surprisingly, bitcoin is lowly correlated to the various indices considered in this study, which is in line with the previous literature (Brière et al., 2015; Bouri et al., 2016; Yermack, 2015; Baur et al., 2018). For example, Brière et al. (2015) compute correlation coefficients over a three-year period and find that bitcoin is lowly correlated with other indices, except for the World inflation-linked bonds and gold. For our seven-year period, bitcoin appears to be weakly correlated even to the bond index and gold.
Table III presents annualised Sharpe ratios over the period 2010-2016 (see also Figures 2, 3, 4 and 5). Bitcoin has the highest Sharpe ratio, 5.09, as compared to 1.11 for S&P 500, for example. Bitcoin recorded its highest ratio in 2013 because of the peak of US$1,000 reached in December 2013. This ratio increased from 2010 to 2013 before decreasing in 2014 (after the closure of Mt. Gox). Afterwards, this ratio increased again from 2014 to 2016. During 2013, the Sharpe ratio of bond, euro, Yen and Yuan was negative.
4.2 Multivariate results
This section estimates the performance of bitcoin by the alpha (risk-adjusted return) of the regression models in equations (1), (2) and (3). Table IV shows the descriptive statistics for the dependent and independent variables. Bitcoin has higher return and higher risk than the global market portfolio and the Fama–French factor-mimicking portfolios. Table V presents the Person correlation coefficients and shows that bitcoin is lowly correlated with these portfolios.
Table VI presents the results of the regression analyses. The bitcoin’s factor loadings are low and insignificant for all the factors considered, and this is true for all regression specifications. The regressions’ intercept, α, is positive and highly significant, with daily values between 0.52 per cent and 0.53 per cent across models and regions (which represents between around 540 per cent and 580 per cent per annum). The Asia-Pacific region provides a slightly higher α than Europe, regardless the model. It can, therefore, be affirmed that investing in bitcoin gives the possibility to earn, both from an economic and statistical perspective, highly significant positive risk-adjusted returns, regardless the region.
Table VII presents the bitcoin’s α by year and by region. α is measured by running the above-specified regressions for each year. Overall, α is relatively higher between 2010 and 2014, and mostly during 2013, with a value of 1.5 per cent per day and exhibits lower values during more recent periods. Over time, the market becoming more efficient and more mature relative to this financial innovation, bitcoin performance has decreased. In particular, informational asymmetries between investors, notably linked to technology understanding, decrease, which may explain this pattern. According to the literature, bitcoin is inefficient on the sub-period 2010-2013 and moves towards efficiency afterwards (Urquhart, 2016; Bartos, 2015). The trend is consistent with the Sharpe ratio trend presented in the previous subsection.
As most bitcoin transactions take place in the Chinese market, we replicate our analyses and estimate α with the three models in which the market, size and B/M factor-mimicking portfolios are composed exclusively of common stocks traded in the Chinese market. These factors are built according to the procedure used by Fama and French used for forming size and B/M portfolios; this procedure is explained in detail on the Kenneth French’s website.
Table VIII presents the correlation coefficients between bitcoin and the explanatory variables used in our regressions in the Chinese market. As for the other regions, bitcoin has a low correlation with the size and B/M factor-mimicking portfolios in China. This result suggests diversification benefits by including bitcoin in a portfolio composed of common stocks in the Chinese market.
The α values estimated with the three performance models on the Chinese market are presented in Table IX. As for the other regions, the α values are economically and statistically highly significant. Despite that the Chinese market is known to be more mature with respect to the bitcoin than other markets (Kajtazi and Moro, 2017), our results show, as for the other regions, strong possibilities to earn positive performance.
5. Robustness checks
5.1 Regression specification
The bitcoin risk-return profile being atypical, it is important to check for problems that may appear with our regressions. Tables XI and XII show that our regressions do not seem to suffer from problems related to multicollinearity (the VIF is always lower than 4), time-series correlation between residuals (the Durbin–Watson statistic is close to the value of 2) and the White’s tests show that the homoscedasticity hypothesis cannot be rejected in most cases.
Another important problem resides in the normality of bitcoin returns. Non-normal returns imply, under the hypothesis that explanatory variables are not stochastic, non-normal errors, in which case the estimators of the regression coefficients are no more efficient. Figures 6 and 7 show that bitcoin returns are skewed to the right, with a skewness of 0.68, and exhibit a “heavy-tailed” distribution, with a kurtosis of 15.98 (see also Table X). Our results are similar to those obtained in previous empirical studies (Baur et al., 2018). Despite that returns look like being normally distributed, normality tests such as Shapiro–Wilk reject the null hypothesis of normality.
We use the residual augmented least squares (RALS) estimators to measure bitcoin performance in a more robust way. This estimation technique is known to provide a more efficient estimation of regression coefficients without imposing specific restriction on the returns’ distribution (Im and Schmidt, 2008; Gallagher and Taylor, 2000). This technique has proved to be useful in analysing, for example, the performance of hedge funds as measured by their α. Heuson and Hutchinson (2015) show that, for such funds, the error generated by regression models depends systematically on skewness. OLS assessment errors prove to be economically significant, and RALS estimation technique proves to be robust to this bias.
We have estimated the RALS α for each one of the three performance models used in our article and for each one of the four regions studied. For expositional convenience, we do not present the procedure in this paper (the reader can refer to Gallagher and Taylor, 2000, amongst others, for a detailed description of this procedure). The qualitative interpretations of our results do not change; they are even reinforced. The RALS α is highly significant both economically and statistically. In the world, Europe, Asia-Pacific and China regions, the RALS α is between 0.52 per cent and 0.59 per cent per day, which is similar to the OLS α . The RALS t-statistic is about 4.43, which is higher than the t-statistic with the OLS procedure (the latter is, on average, of about 3.6).
5.2 Behavioural aspects
Performance results can be controversial if, in reality, the economic value of bitcoin is close to zero. Cheah and Fry (2015) apply econometric modelling on bitcoin prices and find empirical evidence that bitcoin exhibits speculative bubbles and has a fundamental value of zero. However, other authors find that speculation does not explain the high volatility of bitcoin and find evidence that bitcoin has a significantly positive economic value (Blau, 2018; Dwyer, 2015).
The low correlation between bitcoin and asset classes together with the high α obtained in the world, Europe and Asia-Pacific regions may be indicative of a speculative bubble. In this section, we attempt to control for speculative behaviour on bitcoin by using market sentiment as a control variable. Including such control variables is also a way to test the robustness of our results regarding the bitcoin’s performance.
Market sentiment takes into consideration the fact that the efficient market hypothesis does not necessarily apply in the real market. The value of an asset may deviate from its fundamental value because of market’s imperfections, such as information asymmetry under noise, or investors’ behaviour. Market sentiment, or investor sentiment, measures the investors’ behaviour on the market based on their expectations. Prices reveal this expectation, and consequently are an indicator to appraise investor sentiment. If an investor expects that prices are going to increase, thus leading to returns higher than the average, then he/she is bullish and enters on the market with a buyer attitude. On the contrary, a bearish investor expects that prices are going to decrease, and therefore s/he enters the market with a seller attitude. Bullish investors expect positive and greater returns than the fundamental value, whilst bearish investors expect a smaller return on the market.
Based on the literature and especially on the article of Brown and Cliff (2004), we identify four proxies for the market sentiment variable, amongst which three are related to trading and one is related to derivatives, to capture the attitude of investors relative to bitcoin. The first one, named Trade1, is the ratio of advancing issues on declining issues:
The second proxy, Trade2, is based on advancing and declining issues but also on volume, as follows:
The third variable related to the trading group, Trade3, deals with new highs and new lows and captures the relative market strength.
This variable compares the number of stocks reaching new highs and the number of stocks reaching new lows. More precisely, it compares the number of stocks having reached a 52-week high and the number of securities which having hit a 52-week low. A bullish (bearish) sentiment is captured when there are more stocks trading at their highs (lows) than the period before.
Finally, the last variable considered in our analysis is “Put-to-Call” (hereafter PtC) based on the Chicago Board Options Exchange:
A put (call) is the right to sell (buy) an underlying asset at a given price in the future. Buying put (call) means a bearish (bullish) behaviour, whilst selling put (call) means a bullish (bearish) attitude). A bearish (bullish) sentiment is detected when the ratio put/call is high (low).
We add these four variables in the models estimated above and present our results in Tables XIII, XIV and XV. Two of the trade variables, Trade1 and Trade3, are not significant, whilst Trade2 is highly significant. We also note that α is significant in most of the model specifications, except when the derivatives variable (PtC) is considered, in which case α is no more significant in all model specifications. Moreover, Trade2 is negative, whereas we expected a positive sentiment market variable. These results suggest that investors’ behaviour does influence bitcoin’s performance, but overall the performance continues to be positive and significant after controlling for this effect.
Bitcoin is a cryptocurrency that appeared in 2008. The growth of its value exhibits a very volatile price and provides a very specific risk-return profile for investors. After briefly justifying the asset nature of bitcoin, most precisely its common-stock-like nature, our article focuses on its main objective, which is to empirically test its performance. Even if bitcoin belongs to cryptocurrencies, this asset does not seem to respect all properties of a fiat money (medium of exchange, unit of account and store of value). Indeed, its economic and legal characteristics, together with its risk-return profile, make it look mostly like a financial contract. Moreover, its specific risk-return profile leads us to consider that bitcoin may be assimilated to common stock.
Fundamentally, we find that bitcoin is lowly correlated to existing asset classes, which provides opportunities for portfolio diversification, and generates highly significant risk-adjusted returns, of similar magnitude for global, Europe and Asia-Pacific regions. The same result applies for the Chinese market. These results are stable to the specifications of our regression models, which reside in the CAPM, Fama–French three-factor models, and their extensions that includes other factors. Our results are robust to econometric methodologies that account for non-normality problems with the regressions’ errors and to market sentiment variables.
Research could be improved by performing a more robust analysis to test whether bitcoin is a speculative bubble. The high performance obtained by bitcoin in the regions analysed in this paper, and the low correlation with existing assets may be a consequence of this. Furthermore, after controlling for market sentiment with the PtC variable, the bitcoin’s α is no more significant in all model specifications and for all regions. It could be useful in further research to test the market sentiment associated to bitcoin performance by using additional proxies, such as the proxy of closed-end fund which we find in the international market sentiment literature (Lee et al., 1991). Another proxy is the fear and greed Index which takes in consideration different proxies of market sentiment (market momentum, put and call, safe haven demand, stock price breadth and stock price strength, market volatility and junk bond demand) in only one index.
|Variable||N||Mean (%)||SD (%)||Skewness||Kurtosis|
This table presents summary statistics (annualised mean, annualised volatility, skewness and kurtosis) based on daily returns for bitcoin, stock indices (represented by S&P 500 and nasdaq for the USA, FTSE 100 for the UK, DAX30 Germany, NIKKEI225 for Japan and CAC40 for France), MSCI indexes for the world, Europe and Asia-Pacific regions, commodity indexes (gold, oil and commodity), bond index (Pimco) and currencies (dollar, yen, euro and Yuan). The sample is drawn from the blockchain info website and the Datastream database over the period 22 September 2010 to 30 December 2016
Correlation coefficients between bitcoin and various asset classes
|Variable||S&P 500||FTSE 100||DAX30||NIKKEI225||CAC40|
|MSCI World||MSCI Europe||MSCI Asia-Pacific|
This table presents Pearson correlation coefficients between bitcoin returns and asset classes such as stock indices (represented by S&P 500 and nasdaq for the USA, FTSE 100 for the UK, DAX30 for Germany, NIKKEI225 for Japan and CAC40 for France), MSCI indexes for the world, Europe and Asia-Pacific regions, commodity indexes (gold, oil and commodity), bond index (Pimco) and currencies (dollar, yen, euro and Yuan). The sample is drawn from the blockchain info website and the Datastream database over the period 22 September 2010 through 30 December 2016. If the p-value is higher than 5 per cent, then the null hypothesis of no-correlation (rho = 0) is accepted. ***, ** and * indicate that the coefficient is significant at 1%, 5% and 10% level, respectively
This table presents Sharpe ratios per year for Bitcoin, Stock indexes (represented by S&P500 and Nasdaq for the US, FTSE100 for the UK, DAX30 Germany, NIKKEI225 for Japan and CAC40 for France), MSCI indexes for the World, Europe and Asia-Pacific regions, Commodity indexes (Gold, Oil and Commodity), the bond index (Pimco), and cur-rencies (dollar, yen, euro and yuan). The data come from the Datastream database, the Blockchain Info and Fama French websites over the period 22 September 2010 to 30 December 2016 in the World, Europe and Asia-Pacific regions
Descriptive statistics of explanatory variables used in regressions
|Region||N||Mean (/day)||SD||Sum||Min||Max||Mean (/year)||SD (/year)|
This table presents summary statistics (mean, standard deviation (StD), sum, minimum, maximum, annualized mean and annualized volatility) of daily returns expressed in US $ for the dependent variable (RBitcoin-Rf) representing Bitcoin return minus the risk-free rate (proxied by the one-month US Treasury Bill) and for independent variables (RMkt-Rf, which represents the excess return of the market portfolio, SMB, which is the size premium, HML, which is the value premium, bond and gold). The data come from the Datastream database, the Blockchain Info and Fama French websites over the period 22 September 2010 to 30 December 2016 in the World, Europe and Asia-Pacific re-gions. All results are in percentages (%)
This table presents the correlation matrix between the excess return of the Bitcoin (RBitcoin-Rf) and independent variables (RMkt-Rf, SMB, HML, Gold, Bond). The sample is drawn from the Datastream database, the Blockchain Info and Fama-French websites over the period from September 22, 2010, to December 30, 2016, in the World, Europe and Asian-Pacific regions. ***, ** and * indicate that the coeffi-cient is significant at 1%, 5% and 10% level respectively
Performance using the CAPM (Model 1), FF (Model 2) and extensions (Model 3)
|Region||α (%)||RMkt-Rf||SMB||HML||Bonds||Gold||R2 (%)||Annual α (%)|
This table presents regression estimates for the 3 models in equations 1,2,3. The sample is drawn from the Datastream database, the Blockchain Info and Fama-French websites over the period 22 September 2010 to 30 December 2016 in the World, European, and Asia-Pacific regions. ***, ** and * indicate that the coefficient is sig-nificant at 1%, 5% and 10% level respectively
Bitcoin’s α based on Model 2 (Fama and French three-factor model)
|Annualised α (%)||548.75||452.47||1,300.36||810.92||17,076.26||–57.98||189.34||106.43|
|Annualised α (%)||577.23||393.47||1,940.03||914.80||27,716.86||–59.84||193.25||99.73|
|p-value||< 0.01||0.764||0.14||< 0.01||< 0.01||0.25||0.32||0.15|
|Annualised α (%)||583.18||297.41||1,799.39||1,089.78||20,653.39||–62.49||140.02||129.16|
This Table presents the regression results for model 2, year by year, in equation 2. The sample is drawn from the Datastream database, the Blockchain Info and Fama-French websites over the period 2010 2016 in the World, Eu-rope and Asian-Pacific regions. We run year by year model 2 to estimate alpha evolution in the World, European and Asia-Pacific regions. ***, ** and * show that the coefficient is significant at 1%, 5% and 10% level respective-ly
Correlation matrix on the Chinese market
This table presents the correlation matrix between the dependent variable (RBitcoin-Rf) and independent variables (Mkt-Rf, SMB, HML, Gold, Bonds). The sample concerns the period between September 23, 2010, and December 30th, 2016. ***, ** and * show that the coefficient is significant at 1%, 5% and 10% level respectively
Bitcoin’s α on the Chinese market
|Model||α (%/day)||α (%/year)||RMkt-Rf||SMB||HML||Gold||Bonds||R2 (%)|
|p-value||< 0.01||< 0.01|
|p-value||< 0.01||< 0.01||0.78||0.72|
|p-value||< 0.01||< 0.01||0.76||0.70||0.95||0.25|
This table presents the regression results for the 3 models in equations 1,2 and 3 for the Chinese Market. The data needed to construct the proxies for the Chinese market, the size and the B/M indices are extracted from the Datasteam database, while the data needed for computing the Bitcoin’s returns are extracted from Blockchain Info. The period of analysis is from September 23, 2010, and December 30th, 2016. ***, ** and * show that the coefficient is significant at 1%, 5% and 10% level re-spectively
Bitcoin return normality test
|p-value||< 0.01||< 0.01||< 0.01||< 0.01|
This table presents the normality test for Bitcoin returns. The sample is drawn from the Blockchain Info website over the period 2010-2016. The normality hypoth-esis is based on Shapiro-Wilk test: if the p-value is lower than alpha, the null-hypothesis of normality is rejected
Collinearity: test of independent variables
This table shows collinearity test of independent variables. The sample is drawn from the Blockchain Info and Fama-French websites over the period 2010 to 2016 for the World. Europe and Asia-Pacific regions and for the 3 models in equations 1, 2 and 3. VIF is the variance inflation factor: if VIF is higher than 4 and TOL is lower than 0.25, then the variable is considered collinear with the dependent variable
This table presents the residuals analysis from the long-run regressions in the world, Europe and Asia-Pacific regions. The sample is drawn from the blockchain info and Fama-French websites over the period 2010 to 2016 in the world, Europe and Asia-Pacific regions for the three models in equations (1), (2) and (3). The residuals autocorrelation hypothesis is tested using the Durbin-Watson statistic: if the Durbin-Watson statistic is around 2, then the residuals are considered uncorrelated. The homoscedasticity hypothesis is tested based on the White test: if the p-value is lower than 5 per cent, the null hypothesis of homoscedasticity is rejected. The normality hypothesis is based on the Shapiro-Wilk test: if the p-value is lower than 5 per cent, the null hypothesis of normality is rejected
Controlling for market sentiment in the CAPM (Model 1)
|Model||α (%/day)||α (%/year)||RMkt-Rf||Trade1||Trade2||Trade3||Ptc||R2 (%)|
This table controls bitcoin performance by including proxies for market sentiment in the CAPM model (Model 1). We use four proxies of market sentiment: Trade1 is the ratio of advancing issues on declining issues; Trade2 is the ratio of advancing issues on declining issues including volume; Trade3 is the ratio of new highs on news lows; PtC is the ratio of puts to calls. The sample is drawn from the blockchain info website, Datastream, and the NYSE and CBOE website and covers the period between 22 September 2010 and 30 December 2016
Controlling for market sentiment in the FF model (Model 2)
|p-value||< 0.01||0.41||0.55||0.86||< 0.01|
This table controls bitcoin performance by including proxies for market sentiment in the FF model (Model 2). We use four proxies of market sentiment: Trade1 is the ratio of advancing issues on declining issues; Trade2 is the ratio of advancing issues on declining issues including volume; Trade3 is the ratio of new highs on news lows; PtC is the ratio of puts to calls. The sample is drawn from the blockchain info website, Datastream, and the NYSE and CBOE website and covers the period between 22 September 2010 and 30 December 2016
Controlling for market sentiment in the FF extended model (Model 3).
|p-value||< 0.01||0.51||0.37||0.78||0.22||0.46||< 0.01|
This table controls bitcoin performance by including proxies for market sentiment in the extended FF model (Model m 3). We use four proxies of market sentiment: Trade1 is the ratio of advancing issues on declining issues; Trade2 is the ratio of advancing issues on declining issues including volume; Trade3 is the ratio of new highs on news lows; PtC is the ratio of puts to calls. The sample is drawn from the blockchain info website, Datastream, and the NYSE and CBOE website and covers the period between 22 September 2010 and 30 December 2016
DAO is the first “Decentralized Autonomous Organization” which is “an organization operating through a computer program that provides rules of governance to a community. These rules are transparent and immutable because they are written in the blockchain” according to BlockchainInfo. DAO was created in 2016 with the objectives of crowdfunding using the Ethereum platform (the second payment platform which provides both a cryptocurrency, named Ether, and a blockchain platform). In a short period (four weeks), the project involved 20,000 participants and reached US$160m. A failure in the code provoked a huge attack by hackers, which implied the closure of the structure.
Mt. Gox was a bitcoin exchange platform based in Tokyo and created in 2010. It was a very popular platform until the closure of its activities notably linked to bitcoins embezzlement.
Antonopoulos, A. (2015), Mastering Bitcoin, O’Reilly Media.
Bartos, J. (2015), “Does bitcoin follow the hypothesis of efficient market?”, International Journal of Economic Sciences, Vol. IV No. 2, pp. 10-23.
Baur, D.G., Hong, K. and Lee, A.D. (2018), “Virtual currencies: media of exchange or speculative asset?”, Journal of International Financial Markets, Institutions and Money, Vol. 54, pp. 177-189.
Blau, B. (2018), “Price dynamics and speculative trading in bitcoin”, Research in International Business and Finance, Vol. 43, pp. 15-21.
Bouri, E., Moln´ar, P., Azzi, G., Roubaud, D. and Hagfors, L.I. (2016), “On the hedge and safe haven properties of bitcoin: is it really more than a diversifier?”, Finance Research Letters, Vol. 20, pp. 192-198.
Brière, M., Oosterlinck, K. and Szafarz, A. (2015), “Virtual currency, tangible return: portfolio diversification with bitcoins”, Journal of Asset Management, Vol. 16, pp. 365-373.
Brown, G. and Cliff, M. (2004), “Investor sentiment and the near-term stock market”, Journal of Empirical Finance, Vol. 11 No. 1, pp. 1-27.
Burniske, C. and White, A. (2017), Bitcoin: Ringing the Bell for a New Asset Class, Ark Invest Research.
Cheah, E.T. and Fry, J. (2015), “Speculative bubbles in bitcoin markets? An empirical investigation into the fundamental value of bitcoin”, Economics Letters, Vol. 130, pp. 32-36.
Dwyer, G. (2015), “The economics of bitcoin and similar private digital currencies”, Journal of Financial Stability, Vol. 17, pp. 81-91.
Dyhrberg, A.H. (2016a), “Bitcoin, gold and the dollar: a garch volatility analysis”, Finance Research Letters, Vol. 16, pp. 85-92.
Dyhrberg, A.H. (2016b), “Hedging capabilities of bitcoin, is it the virtual gold?”, Finance Research Letters, Vol. 16, pp. 139-144.
Fama, E.F. and French, K.R. (1992), “The cross section of expected stock returns”, The Journal of Finance, Vol. XLVII No. 2, pp. 427-465.
Fama, E.F. and French, K.R. (2012), “Size, value, and momentum in international stock returns”, Journal of Financial Economics, Vol. 105 No. 3, pp. 457-472.
Figuet, J.M. (2016), “Bitcoin et blockchain: quelles opportunités?”, Revue d’économie financière, Vol. 123 No. 3, pp. 325-338.
Gallagher, L.A. and Taylor, M.P. (2000), “Measuring the temporary component of stock prices: robust multivariate analysis”, Economics Letters, Vol. 67 No. 2, pp. 193-200.
Glaser, F., Zimmermann, K., Haferkorn, M., Weber, M.C. and Siering, M. (2014), “Bitcoin – asset or currency? Revealing users’ hidden intentions”, in: Twenty Second European Conference on Information Systems.
Grant, J.M. (2014), Is bitcoin money: implications for bitcoin derivatives regulation and security interest treatment of bitcoins under article 9 of the uniform commercial code, NY University (NYU), School of Law, New York.
Heuson, A.J. and Hutchinson, M.C. (2015), “Which hedge fund managers deliver alpha?” available at: https://ssrn,com/abstract=1786547;http://dx,doi,org/10,2139/ssrn,1786547
Im, K.S. and Schmidt, P. (2008), “More efficient estimation under non-normality when higher moments do not depend on the regressors, using residual augmented least squares”, Journal of Econometrics, Vol. 144 No. 1, pp. 219-233.
Kajtazi, A. and Moro, A. (2017), “Bitcoin, portfolio diversification and Chinese financial Markets”, SSRN Electronic Journal, Vol. 10, available at: http://2139/ssrn,3062064
Kancs, D., Ciaian, P. and Miroslava, R. (2015), “The digital agenda of virtual currencies, can bitcoin become a global currency?” Publications Office of the European Union, JRC Technical report, available at: https://doi.org/10.1007/s10257-016-0304-0
Lakosmki-Laguerre, O. and Desmedt, L. (2015), “L’alternative Monéttaire bitcoin: une perspective institutionnaliste”, Revue de la régulation, Vol. 18, available at: http://journals.openedition.org/regulation/11489;DOI:10.4000/regulation.11489
Lee, C., Shleifer, A. and Thaler, R. (1991), “Sentiment and the closed-end fund puzzle”, The Journal of Finance, Vol. 46 No. 1, pp. 75-109.
Nakamoto, S. (2008), “Bitcoin: a peer-to-peer electronic cash system”, White paper, 9 available at: https://bitcoin.org/bitcoin.pdf
Raiborn, C. and Sivitanides, M. (2015), “Accounting issues related to bitcoins”, Journal of Corporate Accounting and Finance, Vol. 26 No. 2, pp. 25-34.
Selgin, G. (2015), “Synthetic commodity money”, Journal of Financial Stability, Vol. 17, pp. 92-99.
Tschorsch, F. and Scheuermann, B. (2016), “Bitcoin and beyond: a technical survey on decentralized digital currencies”, IEEE Communications Surveys and Tutorials, Vol. 18, pp. 2084-2123.
Urquhart, A. (2016), “The inefficiency of bitcoin”, Economics Letter, Vol. 148, pp. 80-82.
Yermack, D. (2015), “Is bitcoin a real currency? An economic appraisal”, in: Cheun, D.L.K. (Ed.), Handbook of Digital Currency Bitcoin, Innovation, Financial Instruments, and Big Data, chapter 2, pp. 31-43.
Chen, N.F., Kan, R. and Miller, M. (1993), “Are the discounts on closed-end funds a sentiment index?”, The Journal of Finance, Vol. 48 No. 2, pp. 795-800.