国际清算银行-加密交易和比特币价格：来自新零售采用数据库的证据（英）-2022.11-28正式版.docx

《国际清算银行-加密交易和比特币价格：来自新零售采用数据库的证据（英）-2022.11-28正式版.docx》由会员分享，可在线阅读，更多相关《国际清算银行-加密交易和比特币价格：来自新零售采用数据库的证据（英）-2022.11-28正式版.docx（34页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、OBISBIS Working PapersNo 1049Crypto trading and Bitcoin prices: evidence from a new database of retail adoptionby Raphael Auer, Giulio Cornelli, Sebastian Doerr, Jon Frost and Leonardo GambacortaMonetary and Economic DepartmentNovember 2022JEL classification: E42, E51, E58, F31, G28, L50, 032.Keywor

2、ds: Bitcoin, cryptocurrencies, cryptoassets, regulation, decentralised finance, DeFi, retail investment.The largest group of users by far - nearly 40% - were men under the age of 35.While our database does not contain information on the actual performance of the crypto currency investments of indivi

3、duals, we can perform simulations to obtain an estimate. First, we estimate the distribution of the number of users downloading the crypto exchange apps for different Bitcoin prices. We find that 73% of the users downloaded their app when the price of Bitcoin was above $20,000 - above the price of B

4、itcoin in October 2022 (Graph 5). If these users invested in Bitcoin on the same day they downloaded a crypto exchange app, they would have incurred a loss on this initial investment. This compares to 26% of the general population in the countries in our sample - of which 15% are below the age of 19

5、. This finding also mirrors that of Bohr and Bashir (2014), Stix (2019) and Fujiki (2020).8 Men between 35 and 54 made up a further 25% on average. Less than 35% of all users globally are female (Graph 4), and the majority of female crypto app users are under 35. This pattern is consistent with the

6、findings of surveys on cryptocurrency and fintech use; here, too, men are overrepresented (Auer and Tercero-Lucas (2022); Chen et al (2021).”All the young dudes? More than 40% of crypto app users are young menGraph 4In per centWomenLess than 3535-54More than 54Based on active users of 45 crypto exch

7、anges android and iOS apps. Simple averages for the period QI 2020-Q2 2022.Sources: Sensor Tower; authors calculations.Most retail investors downloaded crypto apps when prices were highAlmost three-quarters of users downloaded the app when Bitcoin was higher than $20,000Almost three-quarters of user

8、s downloaded the app when Bitcoin was higher than $20,000Graph 5Percentage of users: Gaining money hh Losing moneyThe graph shows a histogram of the share of daily downloads of crypto-exchange apps by Bitcoin price at the time of first download. Estimations of losses or gains assume that the users p

9、urchased bitcoin on the same daily they downloaded the crypto-exchange app.Sources: CryptoCompare; Sensor Tower; authors, calculations.Second, assuming that each new user bought $100 of Bitcoin in the month of the first app download and in each subsequent month, 81% of users would have lost money (G

10、raph 6). The median investor would have lost $431, corresponding to 48% of their total $900 in funds invested.Only few investors made large gains, while the majority likely lost moneyAssuming an investment of $100 per month, 81% of users have lost moneyGraph 6Percentage of users:Losing money Gaining

11、 moneySources: CryptoCompare; Sensor Tower; authors calculations.3. Empirical analysisBitcoin prices and user numbers have moved in lockstep, with a correlation coefficient of more than 0.9 (Graph 7, left-hand panel). A scatterplot shows that weekly changes in users are correlated with weekly change

12、s in Bitcoin prices (righthand panel), but the relationship is not fully contemporaneous. Indeed, rises in user numbers have lagged rises in prices by an average of about two months.To investigate the relationship between the Bitcoin price and new users in more detail, we estimate variants of the fo

13、llowing regression:= + ,+ +The dependent variable (daily active users or DA Us) is obtained summing the daily numbers of daily active users of these apps at the country level and then taking a monthly average. The result is the monthly average number of daily active users in jurisdiction i for month

14、 t. Our main independent variable is the maximum Bitcoin price in month t, which likely attracts the greatest attention of the investors. We include a set of macro-economic control variables discussed in more detail below. Further, in each specification we include country fixed effects.Table 2 shows

15、 that an increase in the Bitcoin price is associated with a significant increase in the number of new users. On average, a one-percentage point increase in the Bitcoin price is associated with an increase in the monthly average number of daily active users by 1.1% (column 1), significant at the 99%

16、level. This finding doesSimilar price dynamics can be observed for the price of Ether and new users on the Ethereum blockchain (Boissay et al (2022).10Bitcoin prices and user numbers have moved in lockstep, with a correlation coefficient of more than 0.9 (Graph 7, left-hand panel). A scatterplot sho

17、ws that weekly changes in users are correlated with weekly changes in Bitcoin prices (righthand panel), but the relationship is not fully contemporaneous. Indeed, rises in user numbers have lagged rises in prices by an average of about two months.To investigate the relationship between the Bitcoin p

18、rice and new users in more detail, we estimate variants of the following regression:= + ,+ +The dependent variable (daily active users or DA Us) is obtained summing the daily numbers of daily active users of these apps at the country level and then taking a monthly average. The result is the monthly

19、 average number of daily active users in jurisdiction i for month t. Our main independent variable is the maximum Bitcoin price in month t, which likely attracts the greatest attention of the investors. We include a set of macro-economic control variables discussed in more detail below. Further, in

20、each specification we include country fixed effects.Table 2 shows that an increase in the Bitcoin price is associated with a significant increase in the number of new users. On average, a one-percentage point increase in the Bitcoin price is associated with an increase in the monthly average number

21、of daily active users by 1.1% (column 1), significant at the 99% level. This finding doesSimilar price dynamics can be observed for the price of Ether and new users on the Ethereum blockchain (Boissay et al (2022).10 This lagged relationship could suggest that users enter the system attracted by hig

22、h prices and in the expectation that prices continue to rise.Lhs: Bitcoin price (daily)Rhs: Crypto-exchange app daily active users1Graph 7CE) sn4Qo 36U2UChained to speculation? New users enter as the Bitcoin price risesCross-country monthly average of daily active users. Calculated on a sample of mo

23、re than 200 crypto-exchange apps over 95 countries.2 The graph shows a binned scatterplot.Sources: CryptoCompare; Sensor Tower; authors* calculations.not appear to be driven by other financial market or country-specific conditions, as shown in columns (2)-(5). When controlling for stock market retur

24、ns (column 2), turnover (column 3), the global gold price (column 4), or economic policy uncertainty, FX volatility and CPI inflation (column 5), the coefficient on the Bitcoin price remains highly significant and large in magnitude. In our most stringent specification in column (5), a one-percentag

25、e point increase in the Bitcoin price is associated with an increase in new users by 0.9%. These findings suggest that the relation between the entry of the monthly average number of daily active users and the Bitcoin price is not driven by other observable macro-factors.13Crypto adoption rises foll

26、owing increases in the global Bitcoin priceTable 2Dependent variable: ln(monthly average daily active users)Robust standard errors in brackets; */*/* indicates statistical significance at the 1/5/10% level. Regressions include country fixed effects.(I)(ID(UI)(IV)(V)Ln(Bitcoin price)1.109*(0.008)1.07

27、5*(0.008)1.036*(0.009)0.946*(0.013)0.912*(0.012)Ln(MSCI equity index price)1-0.095*(0.022)-0.430*(0.076)-0.271*(0.077)0.058(0.077)Stock market turnover20.304*(0.031)0.249*(0.032)0.185*(0.032)Gold price0.967*(0.085)0.326*(0.092)Global economic policy uncertainty index20.556*(0.041)FX standard deviati

28、on-0.041(0.028)CPI, yoy change0.037*(0.003)Number of observationsR-squared66770.90352600.90747010.90247010.90445160.914Country specific MSC1 equity index price, in local currency. Based on the country specific Datastream equity index, in local currency.Sources: Baker et al (2016); World Bank; Crypto

29、Compare; Datastream; Refinitiv Eikon; Sensor Tower; national data; authors calculations.Differences by user and country characteristicsPrevious literature has established differences in risk tolerance across groups. For example, data from the Survey of Consumer Expectations (SCE) for the United Stat

30、es shows that younger men are more willing to take financial risks than both womenWe additionally control for the network factors identified in Y Liu and A Tsyvinski (2021), namely number of wallets, number of active addresses, number of transactions, number of payments, and the first principal comp

31、onent of these four measures. Overall, our results are robust after controlling for these network factors. We do not report this evidence for conciseness.11and older male respondents (Graph 8). Similar findings have been reported in other contexts (see, for example, Borghans et al, 2009; Arano et al

32、, 2010).To test these arguments, we estimate regressions similar to regression (1), but using the number of new users among different population subgroups as dependent variable.Table 3 shows that young men have a much higher sensitivity to Bitcoin prices than older men or women of any age. The coeff

33、icient on the Bitcoin price is twice as large for young men (column 1) compared to older men (column 2), and more than four times as large relative to women of age 35 and above, which are the least- responsive group. The relationship remains significant at the 99% level for all population groups. Th

34、ese findings suggest that rising prices attract speculative users with a high tolerance for risk. Consistent with this interpretation, recent survey evidence from the UK finds that one of the main reasons for buying cryptocurrencies is “as a gamble that could make or lose moneyM (FCA (2021). Further

35、 analysis (see Graph Al in the appendix) confirms that the stronger reaction of young male users occurs mostly during periods of pronounced price swings.A substantial body of work argues that women tend to be more risk-averse than men (Jianakoplos and Bernasek, 1998). Also on the technology side, th

36、ere are also significant differences in the use of fintech by gender (Chen et al, 2021).12Investigating to what extent the relationship between price development and new users differs across demographic groups could hence offer additional insights. If, for example, risk-seeking segments of the popul

37、ation (ie young men) respond the most to a rising Bitcoin price, this would be consistent with a speculative motive, rather than the search for a safe asset.Willingness to take financial risks for US consumers of age 20-79Index, 1 (lowest willingness)-7 (highest willingness)Graph 84.03.53.02.5IIIIII

38、203040506070AgeFemale MaleWillingness to take financial risks for US consumers of age 20-79. Weighted average (by survey weights) across respondents. The sample covers the period January 2020-July 2021.14Sources: Federal Reserve Bank of New York, Survey of consumer expectations; authors calculations

39、.Table 3Risk aversion: young vs old, male vs female, iOS vs Android usersMonthly average number of users1Male below 35Male above 35Female below 35Female above 35iOSAndroid(D(II)(HI)(IV)(V)(VI)Bitcoin price2.142*(0.137)1.436*(0.091)1.004*(0.062)0.683*(0.042)1.789*(0.109)3.475*(0.223)Number of observa

40、tions838383838383R-squared0.9030.9060.9050.9060.9120.896Robust standard errors in brackets; *7*7* indicates statistical significance at the 1/5/10% level.1 Simple average of the country-level monthly average of DAUs by age and gender. Based on active users of 45 crypto exchanges android and iOS apps

41、.Sources: CryptoCompare; Sensor Tower; authors calculations.(I)(ID(in)(IV)(V)Ln(bitcoin price)0.854*(0.028)0.168*(0.042)0.392*(0.034)1.046*(0.015)1.002*(0.014)Ln(No commercial bank branches per 100k adults)*ln(bitcoin price)0.037*(0.009)Ln(payment apps active users)*ln(bitcoin price)0.046*(0.002)Ln(

42、payment apps downloads)*ln(bitcoin price)0.038*(0.002)Regulatory quality*ln(bitcoin price)-0.137*(0.009)Control of corruption* ln(bitcoin price)-0.107*(0.007)Number of observations44814645464546454645Table 4Dependent variable: ln(monthly average daily active users)R-squared0.9050.9180.9160.9140.914R

43、obust standard errors in brackets; */*/* indicates statistical significance at the 1/5/10% level. Regressions include country fixed effects. Other controls include the natural logarithm of the MSCI equity index price, the stock market turnover, the gold price and the global economic policy uncertain

44、ty index.Sources: Baker et al (2016); CryptoCompare; World Bank; Datastream; Refinitiv Eikon; Sensor Tower; national data; authors calculations.Crypto adoption and institutional characteristicsTaken together, results in Tables 2 and 4 suggest that the Bitcoin price has a positive and highly signific

45、ant association with the entry of new users, even when controlling for other time-varying macro-economic factors or country characteristics. Another way to contrast the relative importance of these different factors is to show how much of the variation in the entry of new users (measured with the R-

46、squared) they can explain.To this end, Graph 9 plots the results from a Shapley decomposition of the R- squared when we regress the number of new users on the Bitcoin price and timevarying macro controls (panel a) or country characteristics (panel b). As shown in panel (a), the Bitcoin price explain

47、s almost 50% of the overall variation in entry of new users, followed by the price of gold with 25%. All other time-varying factors explain less than 15%. Panel (b) shows that these patterns are even more pronounced among country controls. While the price of Bitcoin explains over 55% of the total variation, country characteristics (eg total population, real GDP, the use of payments apps, or institutional quality) explain less than 15% each. These findings suggest that the association between the price of Bitcoin and new users is not only highly sign