[Note: This was originally published on October 20, 2014 at Melotic.com]

Earlier this month CoinDesk published their quarterly State of Bitcoin report.

One stat (on slide 6) that was used to purportedly illustrate growth in user adoption was the increase in the amount of wallets created. According to their figures, there was a 21% increase from June to September this year (5.4 million to 6.5 million respectively).

The problem is however, opening up a wallet or creating a wallet does not constitute growth or adoption, it may simply show interest. Remember: in order to use the Bitcoin network users have to use bitcoins – or more technically speaking, unspent transaction outputs (UTXOs). Wallet creation is a zero-cost economic activity, it is negligible to do so and is not an accurate metric for actually measuring adoption (the use of UTXOs). For a lengthier answer, be sure to read Chapter 4.

On this point, two weeks ago Brian Armstrong, co-founder of Coinbase expressed similar skepticism towards wallet creation numbers, stating: “signups is a poor metric for market share” and had one specific alternative “KYC’ed accounts linked to bank+identity.”

Yet there is a problem with this as well. Both my wife and I have KYC’ed accounts on Coinbase, though neither of us have ever used it (yet). Similarly, I have probably created a handful of wallets over the past year over at Blockchain.info to test out sending several satoshi and never re-used the address. In fact, it is considered ‘best practices’ for end users to not re-use the same address, this is one of the reasons why Electrum and other clients provide multiple addresses to send UTXOs to and from. Another explanation is that some pools switch to dynamic payout addresses which also increases number of new addresses. Thus it should be expected that the amount of “addresses used” or in this case, “wallets created” has increased (recall, there is no such thing as an actual “wallet,” this is just nomenclature to help users better grasp the abstraction that are UTXOs).

Thus, actual bitcoin users (and holders) are probably not a significant fraction of the 6.5 million “wallets” opened through September. That is to say, if a bitcoin user is defined as someone who controls the privkey to a UTXO, it may be the case that bitcoin holders number somewhere between 250,000 to 500,000 globally and has remained in this range for the past six months (see also Android wallet plateau in Chapter 8).

How to measure actual users?

This is an ongoing question, one that has spurred numerous answers. I have written about it at least twice in both books. Some valid metrics include the change in Total Output Volume, Bitcoin Day’s Destroyed and fees to miners. In addition, the Top 500 Richlist is another way to measure on-chain users.

According to their continuously updated Distribution by Address, as of block 320,000 approximately 99.08% of all UTXOs reside on 329,451 addresses. The remaining 0.92% of all satoshis reside on more than 46 million addresses largely in the form of spam, mining rewards, unclaimed tips, etc.

What does this mean?

In April I published a draft of a working paper which used data from block 295,000. At that time, I looked at this slightly differently, noting that 99.08% of used addresses contained less than 1 bitcoin. Andrew Poelstra (andytoshi) corrected this statement, noting that:

“[T]he claim that 99.08% of all addresses contain less than one Bitcoin is an extreme understatement. In fact it is impossible for more than 21 million distinct addresses to correspond to UTXOs containing 1 bitcoin, but there are 1048 addresses. So it will always be the case that at least (100 ‐ 10^ ‐ 38)% of addresses contain less than one bitcoin.”

So what did the actual distribution look like? And what has changed since April? At that time 99.14% of all UTXOs resided on 301,901 addresses. So in the past 5 months there has been a diffusion of less than 1% of those UTXOs to other addresses.

This does not mean there is no activity, or no velocity. Without a full traffic analysis we cannot determine where these UTXOs end up flowing to. Yet it is clear that there has not been a 21% growth in user base during this time frame, otherwise the distribution would have likely changed dramatically (recall that on-chain users cannot participate on the network without at least 5460 satoshi, or the ‘dust limit’ so those marginal holders should not technically be viewed as users).

Again, it is known that certain entities like Bitstamp and Coinbase are large bitcoin holders and they may have on-boarded a number of new users internally. And that some of the addresses containing large amounts of UTXOs likely belong to these types of companies. Yet if there was a 21% growth in the user base over the past 3 months (let alone the 5 month window above), there would likely be other ways to measure and observe this activity as described below.

How to measure adoption?

Due to its pseudonymous nature, one way of measuring adoption is not wallets or price (this largely reflects changes in speculative demand) but in transactional demand. To gauge this metric there are several datum from the blockchain that could be correlated:

• Total Output Volume
• Bitcoin Days Destroyed
• Fees to Miners (or rather “donations”)
• Transactional Volume
• Bitcoin Richlist distribution
• Last usage of a UTXO (as shown by John Ratcliff’s graphics)

In contrast, not a single metric on slide 6 of the CoinDesk report actually measures user adoption. Rather, they all are indicators of interest.

• According to Google Trends, Bitcoin as a term has remained almost flat since this past spring. Perhaps this will change, but interest is not the same as adoption.
• Hashrate is not an accurate measure of user growth or adoption as it measures hashrate not usage (and in fact, the amount of actual miners has likely decreased since the advent of ASICs).
• Github repos may potentially be an accurate if these updates/requests were substantial changes, yet as I have explained elsewhere – most of the innovation has been outsourced to altchains which can afford experimentation (e.g., smaller community, less impact if it fails). Instead, changes to the core protocol are relatively slow and conservative (understandably). To quote Chapter 4:

Similarly, if a serious flaw and vulnerability was found in the core Bitcoin code base (bitcoind) which caused a cascade of hard forks that destroyed Bitcoin entirely, the github commit component would precisely measure the wrong thing, inputs, rather than an accurate attribute: healthy production code. In fact, that measure would spike, leading observers to believe that this collapse is good news for Bitcoin.

• In terms of merchant adoption, while this has indeed increased, merchants are still dependent on a fixed slice of liquidity (roughly 10% of all mined coins are liquid in any given month). Furthermore, because there is no “circular flow of income,” the vast majority of coins are usually immediately converted back into fiat. Thus, again, merchant adoption should not be conflated with user adoption.
• VC funding is an indicator of interest and changes in sentiment, but not necessarily growth or adoption (unless these VC deals are done in bitcoin, which some of them are, perhaps this will increase in time).
• While ATMs will likely continue to be purchased, built and installed, it is unclear at this time if there will be non-marginal growth from these on/off ramps. ATM owners have overhead costs (amortizing machine costs, maintaining a physical presence and compliance), costs that may be added onto the end user and perhaps lowering the demand (due to price elasticities) in certain regions and corridors. These changes in demand could be viewable on-chain through the metrics above.
• Lastly, “Merchant’s annual revenue” is misleading because that is unrelated to how much revenue they generate from digital currencies. Perhaps digital currencies will eventually impact the bottom line, but total revenue is not a reflection in user adoption of digital currencies.

While future posts may look into these slides more it is necessary to point out that these interest metrics above could turn into user adoption (depends on what the “bitcoin sales cycle” turns out to be).

Readers are also encouraged to look through Sarah Meiklejohn’s research on this topic entitled, A Fistful of Bitcoins: Characterizing Payments Among. Men with No Names. Combing through and correlating this type of data may also be a good research project for students this fall and winter.