Full Terms digital currency; Google search data; Libertarians;illegal activityJEL Classification E42; F33; K42; K49I. IntroductionBitcoin, a virtual global currency, has been the topicof much media, Internet and policy discussion. Over13.4 million Bitcoins are in circulation and have atotal market value of 4.6 billion.1Little is knownabout the characteristics of Bitcoin users, eventhough thousands of businesses accept Bitcoins aspayment. Transactions with Bitcoin are near anon-ymous due to the cost associated with identifying auser’s electronic signature. Although some conveni-ence sampling exists of Bitcoin enthusiasts, no sys-tematic data collection has been done.We use Google Trends hereafter, ‘GT’datatostudy the clientele driving interest in Bitcoin, withthe caveat that search query interest need not implyactive participation. Based on anecdotal evidenceabout Bitcoin users, we construct proxies for fourpossible clientelecomputerprogramming enthusiasts,speculative investors, Libertarians and criminals.Illegal activity and computer programming are bothpositively associated with Bitcoin use, while no asso-ciation exists for Libertarian ideology or investmentmotives in most specifications.II. The Bitcoin MarketBitcoin was created in 2009 as an unregulated, alter-native of exchange for online payments.Upon signing up for an account, an individualreceives an electronic signature that secures transac-tions and disallows double spending enforced by adiverse computer network. This process circum-vents conventional s that involve trust inand fees to a third party. Conventional sinvolve third-party fees, deterring small transactionsNakamoto, 2008.2Anonymity is theoretically*Corresponding author. E-mail aaronuky.edu1https//blockchain.info/charts/total-bitcoins2https//bitcoin.org/bitcoin.pdfApplied Economics Letters, 2015Vol. 22, No. 13, 1030–1036, http//dx.doi.org/10.1080/13504851.2014.9953591030 2015 Taylor ‘miners’ the term for those seekingto discover new Bitcoins can earn the currency inexchange for utilizing special software to authenti-cate real-time Bitcoin transactions.4The anonymityof Bitcoin is attractive for criminal activity. The 2October 2013 FBI takedown of the Silk Road web-site – an online marketplace ‘for everything fromheroin to forged passports’ where transactions tookplace in Bitcoins – highlighted the importance ofBitcoin’s perceived anonymity and led to a 22reduction in Bitcoin’s price.5In order to understand the underlying rationale forBitcoin use, Lui 2013 surveyed 1133 members ofthe Bitcoin community by posting links on Bitcoinwebsites.6The survey identified three key motivescuriosity, profit and political. Respondents whichincluded both owners and nonowners of Bitcoinare likely unrepresentative of the larger community;for example, those using Bitcoin for illegal activityare unlikely to participate.IV. GT DataWe collected GTsearch query data fromJanuary 2011to July 2013 for all US states and Washington DC.7We lookedforterms related to Bitcoin and its possibleclientele.8Some of these correlations are inherentlydifficult to measure, due to the sensitivity of the activ-ity; Stephens-Davidowitz 2013, 2014 argues, how-ever, that Google data are unlikely to suffer frommajor social censoring, and uses GT to explore childabuse and racial animus.9Although it is conceivablethat higher Bitcoin search volume need not translateinto increased market participation, Kristoufek 2013demonstrates a strong positive correlation betweenBitcoin searches and exchange prices.GT can be used to extract data for precise searchterms and more general topics see Fig. 1. Searchtermswillreturndatafortheexactquerywhiletopicscount related searches too.10For instance, the topic‘Bitcoin Currency’ includes the terms ‘Bitcoin’,‘Bitcoins’, ‘Bitcoin Mining’, ‘Bit Coin’, ‘Bitcoinexchange’, ‘Bitcoin price’ and ‘Bitcoin value’.Weuse search topics for Bitcoin under ‘Currency’ andComputer Science under ‘Discipline’. For otherclienteles – Illegal Activity, Libertarians andSpeculative Investors – we use the search terms‘Silk Road’, ‘Free Market’ and ‘Make Money’,respectively.11GT does not report raw search counts for a topic;such counts would be misleading because Google’s3http// andhttps// inJanuary2011because GTbetter measures state-levelsearchactivityfromthat point.WeendedinJuly2013because the ‘Silk Road’ website – unknown to most of the public – was shut down soon after and made front-pageheadlines in national publications.8GT data have been predictive of behaviour in diverse economic markets including entertainment, labour and housingAskitas and Zimmerman, 2009; Varian and Choi, 2009; Hand and Judge, 2012; Wu and Brynjolfsson, 2013. It has alsobeen used for detecting health patterns, including influenza outbreaks and Lyme disease cycles Ginsberg et al., 2009;Carneiro and Mylonakis, 2009; Seifter et al., 2010.9Heshowsthatcross-sectionalstatevariationinGTishighlycorrelatedwithotherdatasources;forexamplethesearchratefor the word ‘God’ explains 65 of the variation in the percentage of a state’s residents believing in God.10https// attempted to use alternative terms for these concepts such as ‘Libertarian’ or ‘Ron Paul’ for Libertarianism, butsearch interest was either too sparse or had a strong political cycle.Characteristics of Bitcoin users an analysis of Google search data 1031Downloaded by [University of Kentucky Libraries] at 0626 27 November 2015 popularity and search queries grow over time.12Instead GT computes the number of topic searchesrelative to all searches, normalizes the series so thehighest value is 100, and scales all other valuesrelative to the highest. Figure 2 illustrates theBitcoin time series in California, where popularitypeaked in April 2013. For each state, we initiallycompute a 31-month time series for the relativepopularity of Bitcoin and each clientele grouping.13We thenuseGTto measurerelative state-level popu-larityofeachsearchtermforthefullperiodandscaleeach state-series relative to the most popular state.During the observed time frame, the states with thehighest interest in Bitcoin were Utah, Oregon,California, Washington, Nevada, New Hampshireand Vermont see Fig. 3 We then rescale eachstate-specific time series by its geographicpopularity. Thus, using California’s value of 94Fig. 1. Google ‘search term’ versus ‘topic Currency’. Google and the Google logo are registered trademarks ofGoogle Inc., used with permissionSource Google Trends states and search terms had weekly activity such as California’s Bitcoin activity in Fig. 2. In such cases, wecomputed monthly averages for all nonmissing values and then rescaled the series with a maximum value of 100.1032 A. Yelowitz and M. WilsonDownloaded by [University of Kentucky Libraries] at 0626 27 November 2015 from the geographic Bitcoin comparison, the entireCaliforniatimeserieswouldberescaledto0.94ofitsoriginal value.Our outlined ology presents us with twolimitations. First, GT samples its database and com-putes the index based on that sample.14Weobserved slightly different values for the index byrefreshing the web page, even with the samerestrictions. Although the overall conclusions areunlikely to change from sampling, this prohibitsexact replication. Second, GT gives a value ofzero if it cannot gather enough data.15We excludestate-month observations with missing values.While every index has missing values for particularmonths, some states returned a missing value in thecross-sectional analysis, which prevents rescalingof the state-specific time series. Delaware, NorthDakota, and Wyoming were excluded as they hadmissing values for ‘Free Market’ and/or ‘SilkRoad.’ Out of 1488 48 states 31 months poten-tial observations, our analysis uses 794 with non-missing values on Bitcoin, Computer Science, FreeMarket, Silk Road and Make Money. The mostpopulous states tend to have the fewest missingstate-month observations.V. Empirical ResultsFollowing Stephens-Davidowitz 2014, we normal-ize each search rate to its z-score and estimate thefollowing specificationFig. 2. Index for Bitcoin topic search California time series, January 2011–July 2013. Google and the Google logoare registered trademarks of Google Inc., used with permission14https// of Bitcoin users an analysis of Google search data 1033Downloaded by [University of Kentucky Libraries] at 0626 27 November 2015 BITCOINjt β0β1Xjtδjδtεjt1where BITCOINjtis Bitcoin interest in state j inmonth t, Xjtis clientele interest, and δjand δtarestate and time fixed effects. Each state-month isweighted by state population in July 2011, and SEsare corrected for non-nested two-way clustering atthe state and time levels Cameron et al., 2011. Byincluding fixed effects in our fully saturated specifi-cation, the impact of clientele association on Bitcoinismeasuredthroughdifferentialwithin-statechangesover time Yelowitz, 1995.Resultsforavarietyofspecificationsarepresentedin Table 1, Columns 1–3 progressively includeadditional controls for state and time. The inclusionof both state and time fixed effects identifies interestin Bitcoin by exploiting within-state changes overtime. In this specification, interest in computerscience and Silk Road is both positively associatedwith interest in Bitcoin and is statistically significantat the 10 level. The interpretation of the specifica-tion in column 3 is the following a one-SDincrease in computer science interest leads to a 0.13SD increase in Bitcoin interest, while a one-SDincrease in Silk Road interest leads to a 0.09 SDincrease in Bitcoin interest. Column 4 adds a ‘pla-ceboclientele’–searchesforthesingerMileyCyrus.Reassuringly, inclusion of this placebo variableneither changes any of the inferences on the otherclientele, nor is the variable itself significant.Columns 5–6 interact each clientele searchterm with average monthly Bitcoin prices. Profit-motivated clientele – such as speculative investors– may find Bitcoin more intriguing when prices arehigh. However, we again observe a positiveFig. 3. Index for Bitcoin topic search cross-sectional popularity, January 2011–July 2013. Google and the Googlelogo are registered trademarks of Google Inc., used with permission1034 A. Yelowitz and M. WilsonDownloaded by [University of Kentucky Libraries] at 0626 27 November 2015 Table1.DeterminantsofBitcoinsearchinterest123456789101112ComputerScience0.0830.1430.1250.1240.0090.0080.1210.1210.0110.1310.0140.1250.0660.1730.0730.0730.0290.0280.0590.0590.0270.0640.0300.065ComputerScience0.2080.2090.2050.202XPRICE/1000.0680.0680.0640.062SilkRoad0.9481.0800.0930.093−0.007−0.0070.0760.076−0.0120.1050.0100.0880.3740.4080.0510.0520.0400.0400.0390.0400.0360.0660.0380.044SilkRoad0.1930.1920.1850.141XPRICE/1000.1010.1000.0970.082FreeMarket0.211−0.1720.0230.023−0.006−0.0050.0310.0310.0030.0360.0040.0210.0760.0580.0220.0220.0220.0210.0190.0190.0190.0250.0210.020FreeMarketXPRICE/1000.0430.0470.030−0.0110.0680.0800.0770.073MakeMoney0.0520.085−0.004−0.0040.0040.0040.0050.0050.016−0.0410.0030.0060.0890.1210.0260.0260.0260.0250.0300.0290.0260.0470.0290.030MakeMoneyXPRICE/100−0.039−0.045−0.069−0.0950.0700.0750.0760.075MileyCyrus0.0210.0310.0150.0340.0400.0800.0400.075MileyCyrus0.0100.007XPRICE/1000.1150.101Unemp.Rate−0.1210.064−0.1210.064−0.0800.051−0.2810.097−0.2030.072−0.1050.063NotesSamplesizeis794incolumns1–9,591incolumns10and112012onwardsand580incolumn12stateswith≥20observations.SEscorrectedfornon-nested,two-wayclusteringattheSTATEandMONTHlevels.Observationsweightedbypopulation.Stateandtimefixedeffectsincludedincolumns3–12.Statefixedeffectsandatimetrendincludedincolumn2.Characteristics of Bitcoin users an analysis of Google search data 1035Downloaded by [University of Kentucky Libraries] at 0626 27 November 2015 association between Bitcoin interest and our twoclientele groups of computer programming enthu-siasts and those possibly engaged in illegal activityin the interaction term, not the main effect. Theother clientele groups remain insignificant.Columns 7–9 include the state-level monthlyunemployment rate. Columns 7–8 show that theinferences on computer science and illegal activityare unchanged, but there is some evidence thatLibertarian activity also drives interest in Bitcoinalthough the specification including interactionswith Bitcoin prices is insignificant. Higher unem-ployment rates are negatively associated withBitcoin interest. Columns 10–11 estimate themodel from 2012 onwards when Bitcoin was morepopular, while column 12 estimates it for the 24states with at least 20 monthly observations. In allcases, fluctuations in computer science and illegalactivity continue to drive Bitcoin interest, as well asthe business cycle.VI. DiscussionAlthough many commentators have speculatedabout motives for using Bitcoin, our study is thefirst to systematically analyse Bitcoin interest,including the interest of hard-to-observe clientele.We findrobustevidencethatcomputerprogrammingenthusiasts and illegal activity drive interest inBitcoin and find limited or no support for politicaland investment motives.ReferencesAskitas, N. and Zimmerman, K. F. 2009 Google econo-metrics and unemployment forecasting, AppliedEconomics Quarterly, 55, 107–20. doi10.3790/aeq.55.2.107Cameron, A. C., Gelbach, J. B. and Miller, D. L. 2011Robustinferencewithmultiwayclustering,JournalofBusiness Economic Statistics, 29,238–49.doi10.1198/jbes.2010.07136Carneiro, H. A. and Mylonakis, E. 2009 Google Trends aweb-based tool for real-time surveillance of diseaseoutbreaks, Clinical Infectious Diseases, 49,1557–64.doi10.1086/630200Ginsberg, J., Mohebbi, M. H., Patel, R. S., et al. 2009Detecting influenza epidemics using search enginequery data, Nature, 457, 1012–14. doi10.1038/nature07634Hand, C. and Judge, G. 2012 Searching for thepicture forecasting UK cinema admissions usingGoogle Trends data, Applied Economics Letters,19,1051–55. doi10.1080/13504851.2011.613744Kristoufek, L. 2013 BitCoin meets Google Trends andWikipedia quantifying the relationship betweenphenomena of the Internet era, Scientific Reports, 3,1–7.