I created dormantbitcoin.com, in particular, dormant bitcoin (DB) chart, so the picture of the dormancy of all bitcoins becomes clear.
A bitcoin address is called dormant since its last sending block. Accordingly, all its bitcoins are said to be stayed in that block. In the Dormant Bitcoin (DB) chart, the X-axis is the block range, and the Y-axis is the number of bitcoins staying in the corresponding block range.
Topic: How to determine dormant coins? (Inspired by Shamir's paper) (Read 5362 times)
Instead if Bitcoin days destroyed, how about a total of Bitcoin days not destroyed?
One bitcoin day not destroyed (or just One Bitcoin Day) is 1 BTC that has been received one day ago. Five Bitcoin Days is 5 BTC that has been received one day ago, or 1 BTC received five days ago.
For all 10M or so bitcoins, you check when they have been sent to the address they are at now and add it together.
If 78% of all bitcoins are dormant and the rest is used randomly, the number of Bitcoin Days should rise by ( .78 * numBitcoinsInExistence ) per day.
If, suddenly, all bitcoins were used on the same day, Bitcoin Days would drop to 0.
You could measure the dormancy of bitcoins over a period with the following formula:
Which would give you a number > 1 if bitcoins became more dormant, 1 > number > 0 if the coins circulating became more circulated, or number < 0 if previously dormant numbers came back in circulation.
One bitcoin day not destroyed (or just One Bitcoin Day) is 1 BTC that has been received one day ago. Five Bitcoin Days is 5 BTC that has been received one day ago, or 1 BTC received five days ago.
For all 10M or so bitcoins, you check when they have been sent to the address they are at now and add it together.
If 78% of all bitcoins are dormant and the rest is used randomly, the number of Bitcoin Days should rise by ( .78 * numBitcoinsInExistence ) per day.
If, suddenly, all bitcoins were used on the same day, Bitcoin Days would drop to 0.
You could measure the dormancy of bitcoins over a period with the following formula:
Code:
( NumBitcoinDaysByDate(endDate) - NumBitcoinDaysByDate(startDate) ) / numBitcoinsByDate(startDate) / numDaysBetween(startDate, endDate)
it's like trying to determine money velocity of the paper dollar under the gold exchange standard by studying the movements of gold between central banks
You pretty much nailed it right there.
Mt. Gox has twice now released "transparency" updates in which they share the amount of fiat deposits they received in the prior 30-day period.
This is probably the most useful metric in trying to come up with money flows into bitcoin. But once that value is transferred into bitcoin, the difference between a pseudonymous digital currency and the traditional monetary systems becomes glaring.
Aren't Bitcoin Days Destroyed a good enough quantitative measure?
The problem with this thread is that it was started based on the assumption that there get to be a method to determine quantitatively the proporition of "dormant" coins.
What if this idea doesn't make sense?
What is a "dormant" coin after all?
Assuming you get a different result with any value of T, where T is the period during which an address has to retain its balance to qualify as "dormant".
Then when is the right T? why not a few more days? Or a few less weeks?
This "dormant" criterion is like trying to decide when you have got enough sand to have a heap of sand.
IMO, the analysis need to be done and ploted on many evenly spaced values of T, or none.
And some general conclusion on the past behavior can only be made if there is consistent trend in the results.
Picking a value at random does not allow to make general conclusions.
Beside, I am a bit warry about this quote:
I'd like to highlight that should the community reach a consensus about a reasonable way to calculate this metric as requested by Dr. Shamir, it has to be clear that this is only that and nothing else: a measure of the number of coins that haven't moved within a given timeframe.
There is absolutely no basis to deciding that this measure represents savings, or hoarding or whatever.
Which means there is no ground for claiming this could affect anyone's political agenda, as suggested in this quote.
Regardless of the results found, I urge Dr. Shamir to refrain from making connections in his revised paper that aren't backed by sound evidence.
I am going to give an example that, hopefully, will help understand why coin inertia cannot in anyway be related to money velocity in the economy.
Many people use online wallets, which also includes balances at exchanges. Due to the networking effect, people will tend to subscribe to the same wallet or exchange service as other people with whom they wish to transact on a regular basis. Which means that a significant number of transactions will happen within wallet and exchange systems. BUT, and this is very important; within a given exchange or wallet system, transactions are done using the system internal ledger and do not appear in the blockchain as they do not require actual bitcoin transfers. Since actual coin deposits and withdrawals are only a fraction of the exchange and wallet sites activity, exchanges and wallets tend to put most of their coin deposits in cold storage. Depending on whether the operator uses a FIFO (First In First Out) or LIFO (Last In First Out) approach to managing cold storage, large amounts of coins could appear to be "dormant" as long as there isn't a major bank run.
Why does the above invalidates the proposed study? Because there is a very weak and loosy relation between deposit / withdrawals, and money spending. For the sake of the example, MtGox could have a daily turnover of half of the market cap, and involving several times the total amount of bitcoins on deposit, and yet not write a single transaction in the blockchain.
By analogy, it's like trying to determine money velocity of the paper dollar under the gold exchange standard by studying the movements of gold between central banks, and totally dismissing that the economy was in fact using paper dollars, and central banks making massive reserves in foreign currencies and book keeping, and rebalancing gold reserves only occasionaly if at all. By applying the same methodology as in the proposed paper, one would find that 95% of the gold-backed dollars were dormant, and that the world economy was on the verge of clinical death for about half a century...
In short, I think this measurement doesn't make any economic sense.
Now, this isn't a paper about economy, but about quantitative measurenents.
Therefore I am fine with whatever measure are published *so long as no claim is made that these figures have any economic significance, or represent anything else but what they are meant to be: quantitative measures of some sort*.
The problem with this thread is that it was started based on the assumption that there get to be a method to determine quantitatively the proporition of "dormant" coins.
What if this idea doesn't make sense?
What is a "dormant" coin after all?
Assuming you get a different result with any value of T, where T is the period during which an address has to retain its balance to qualify as "dormant".
Then when is the right T? why not a few more days? Or a few less weeks?
This "dormant" criterion is like trying to decide when you have got enough sand to have a heap of sand.
IMO, the analysis need to be done and ploted on many evenly spaced values of T, or none.
And some general conclusion on the past behavior can only be made if there is consistent trend in the results.
Picking a value at random does not allow to make general conclusions.
Beside, I am a bit warry about this quote:
Quote from: Adi Shamir
However, there is one thing I would like toclarify: I do not want to have a target-shopping experiment, in which once you will see the results you will decide whether you like it or would prefer to change the methodology again in order to geta politically more appealing result. We thus propose to wait until some consensus emerges, and only then run the selected methodology on the data and announce the results.
I'd like to highlight that should the community reach a consensus about a reasonable way to calculate this metric as requested by Dr. Shamir, it has to be clear that this is only that and nothing else: a measure of the number of coins that haven't moved within a given timeframe.
There is absolutely no basis to deciding that this measure represents savings, or hoarding or whatever.
Which means there is no ground for claiming this could affect anyone's political agenda, as suggested in this quote.
Regardless of the results found, I urge Dr. Shamir to refrain from making connections in his revised paper that aren't backed by sound evidence.
I am going to give an example that, hopefully, will help understand why coin inertia cannot in anyway be related to money velocity in the economy.
Many people use online wallets, which also includes balances at exchanges. Due to the networking effect, people will tend to subscribe to the same wallet or exchange service as other people with whom they wish to transact on a regular basis. Which means that a significant number of transactions will happen within wallet and exchange systems. BUT, and this is very important; within a given exchange or wallet system, transactions are done using the system internal ledger and do not appear in the blockchain as they do not require actual bitcoin transfers. Since actual coin deposits and withdrawals are only a fraction of the exchange and wallet sites activity, exchanges and wallets tend to put most of their coin deposits in cold storage. Depending on whether the operator uses a FIFO (First In First Out) or LIFO (Last In First Out) approach to managing cold storage, large amounts of coins could appear to be "dormant" as long as there isn't a major bank run.
Why does the above invalidates the proposed study? Because there is a very weak and loosy relation between deposit / withdrawals, and money spending. For the sake of the example, MtGox could have a daily turnover of half of the market cap, and involving several times the total amount of bitcoins on deposit, and yet not write a single transaction in the blockchain.
By analogy, it's like trying to determine money velocity of the paper dollar under the gold exchange standard by studying the movements of gold between central banks, and totally dismissing that the economy was in fact using paper dollars, and central banks making massive reserves in foreign currencies and book keeping, and rebalancing gold reserves only occasionaly if at all. By applying the same methodology as in the proposed paper, one would find that 95% of the gold-backed dollars were dormant, and that the world economy was on the verge of clinical death for about half a century...
In short, I think this measurement doesn't make any economic sense.
Now, this isn't a paper about economy, but about quantitative measurenents.
Therefore I am fine with whatever measure are published *so long as no claim is made that these figures have any economic significance, or represent anything else but what they are meant to be: quantitative measures of some sort*.
I sent the authors this. I do agree with gmaxwell in general, but at least this metric gives a reasonable lower bound under the same assumptions they were using. (IE, in reality it would easily not be a lower bound for cold stored coins by exchanges)
Regarding my point on bank balances, if anyone has access to data from actual banks I'd be very curious to see that question answered.
Regarding my point on bank balances, if anyone has access to data from actual banks I'd be very curious to see that question answered.
Quote
Hi Dorit and Adi,
I read your recent paper on Bitcoin. On the topic of quantifying coin storage vs. movement I'd suggest thinking about making use of price data for currencies Bitcoin can be converted to. Specifically to distinguish between coins that haven't been moved because they are in savings accounts, vs. coins that haven't been moved because the private keys have been lost, I'm suggesting making the lower-bound assumption that any coin lost was lost immediately after the last transaction to it. Thus to the owner at the time of this hypothetical key loss the value of the private key to the owner was the value in USD that the coins could have been converted to at this time. This method would discount the value of coins lost during the initial block chain history where many believe that the corresponding keys are simply no longer exist. Simply counting all the coins that haven't been moved recently and multiplying that number by the BTC/USD exchange rate then provides an upper bound.
USD price data from the Mt. Gox exchange is available since July 18th 2010 when the exchange opened.(1) Other ways of buying Bitcoins before that time did exist, such as NewLibertyStandard, although as far as I know trading volume and prices were extremely low.(2) Simply assuming a price of $0 before the opening of Mt. Gox isn't unreasonable.
Of course there still are issues with this measurement. As an example consider how many entities holding Bitcoins for others will keep the vast majority of the coins held off line. (3) Thus if incoming and outgoing transactions balance the movement of the bulk of the coins will be very small, while the velocity of a small groups of coins will be quite high; I note that you say coins in circulation have been moved a much larger than expected number of times. This of course is identical to how banks in the wider world work, except that in the wider world banks have the opportunity to lend deposits out, whereas currently the market for loans denominated in Bitcoins isn't mature enough to allow this. (and quite possibly the depositors wouldn't accept such loans anyway (4)) Of course an interesting comparison to make would be the percentage of "untouched" balances for your average bank. While not accurate in what actually happens to the funds, it would be an accurate analogy to Bitcoin savings from the account holders perspective. The movement of gold held is similar situation.
You should also look into the related idea of bitcoin days destroyed: https://en.bitcoin.it/wiki/Bitcoin_Days_Destroyed
More generally applying price in the discussion around anything Bitcoin related is important. For instance the discussion of the large transactions needs to take into account that the Nov 8th transaction, 90,000BTC, represents only 22.5K USD of value while more recent ~500k large transactions represent over 250 times more value transacted.
(1) https://docs.google.com/spreadsheet/ccc?key=0AmcTCtjBoRWUdHJuUE1mUkFxa3A0eHBDQkxZLVVFZmc#gid=0 It seems that their data source is missing the past month; you might want to ask bitcoincharts.com for an up-to-date copy.
(2) http://newlibertystandard.wetpaint.com/page/2009+Exchange+Rate
(3) Example: http://blog.coinbase.com/post/33197656699/coinbase-now-storing-87-of-customer-funds-offline They claim the amount held offline is 90%, with the largest transaction in a given weak about 5% of funds held.
(4) http://support.coinbase.com/customer/portal/articles/667213-do-you-keep-fractional-reserves-of-customer-deposits-
I read your recent paper on Bitcoin. On the topic of quantifying coin storage vs. movement I'd suggest thinking about making use of price data for currencies Bitcoin can be converted to. Specifically to distinguish between coins that haven't been moved because they are in savings accounts, vs. coins that haven't been moved because the private keys have been lost, I'm suggesting making the lower-bound assumption that any coin lost was lost immediately after the last transaction to it. Thus to the owner at the time of this hypothetical key loss the value of the private key to the owner was the value in USD that the coins could have been converted to at this time. This method would discount the value of coins lost during the initial block chain history where many believe that the corresponding keys are simply no longer exist. Simply counting all the coins that haven't been moved recently and multiplying that number by the BTC/USD exchange rate then provides an upper bound.
USD price data from the Mt. Gox exchange is available since July 18th 2010 when the exchange opened.(1) Other ways of buying Bitcoins before that time did exist, such as NewLibertyStandard, although as far as I know trading volume and prices were extremely low.(2) Simply assuming a price of $0 before the opening of Mt. Gox isn't unreasonable.
Of course there still are issues with this measurement. As an example consider how many entities holding Bitcoins for others will keep the vast majority of the coins held off line. (3) Thus if incoming and outgoing transactions balance the movement of the bulk of the coins will be very small, while the velocity of a small groups of coins will be quite high; I note that you say coins in circulation have been moved a much larger than expected number of times. This of course is identical to how banks in the wider world work, except that in the wider world banks have the opportunity to lend deposits out, whereas currently the market for loans denominated in Bitcoins isn't mature enough to allow this. (and quite possibly the depositors wouldn't accept such loans anyway (4)) Of course an interesting comparison to make would be the percentage of "untouched" balances for your average bank. While not accurate in what actually happens to the funds, it would be an accurate analogy to Bitcoin savings from the account holders perspective. The movement of gold held is similar situation.
You should also look into the related idea of bitcoin days destroyed: https://en.bitcoin.it/wiki/Bitcoin_Days_Destroyed
More generally applying price in the discussion around anything Bitcoin related is important. For instance the discussion of the large transactions needs to take into account that the Nov 8th transaction, 90,000BTC, represents only 22.5K USD of value while more recent ~500k large transactions represent over 250 times more value transacted.
(1) https://docs.google.com/spreadsheet/ccc?key=0AmcTCtjBoRWUdHJuUE1mUkFxa3A0eHBDQkxZLVVFZmc#gid=0 It seems that their data source is missing the past month; you might want to ask bitcoincharts.com for an up-to-date copy.
(2) http://newlibertystandard.wetpaint.com/page/2009+Exchange+Rate
(3) Example: http://blog.coinbase.com/post/33197656699/coinbase-now-storing-87-of-customer-funds-offline They claim the amount held offline is 90%, with the largest transaction in a given weak about 5% of funds held.
(4) http://support.coinbase.com/customer/portal/articles/667213-do-you-keep-fractional-reserves-of-customer-deposits-
What would be the best way to measure how many coins are in long-term savings, as opposed to being actively circulated?
Presumably if large balances that are held at EWallet providers are then withdrawn (e.g., sold at the exchange and withdrawn by the buyer), they would then need to come from the offline / cold wallet storage. That activity would likely appear in the BitcoinDays destroyed metric as any funds in the offline / cold wallet storage would likely be older. If the EWallet provider where this is happening has some manual method where coin selection follows last-in-first-out or first-in-first-out, and whatever convention is used is known, then that might help increase the accuracy of BitcoinDays destroyed, but I don't believe that info is shared.
- http://en.bitcoin.it/wiki/Bitcoin_Days_Destroyed
- https://bitcointalksearch.org/topic/m.90789
What would be the best way to measure how many coins are in long-term savings, as opposed to being actively circulated?
Which of your (1), (2), (3) in this case covers coins which are deposited in a bank like service but are actively changing ownership inside it with no blockchain activity?I'm not aware of any method which will give reliable results and, worse, I'm not aware of any method which will give results where we can give a reliable distribution of the errors.
Exactly. Worse than having no information at all is having bad information and it giving a false sense of security when used. With EWallets, using the blockchain for gleaning information about bitcoin distribution won't even be in the same ballbpark.
What would be the best way to measure how many coins are in long-term savings, as opposed to being actively circulated?
Which of your (1), (2), (3) in this case covers coins which are deposited in a bank like service but are actively changing ownership inside it with no blockchain activity?I'm not aware of any method which will give reliable results and, worse, I'm not aware of any method which will give results where we can give a reliable distribution of the errors. The errors could enormously be in one direction or another based on the levels of hidden activity, proper non-reuse of addresses, and self mixing for consolidation/bookeeping and traffic analysis frustration. The levels of these cofounders are unmeasurable.
What would be the best way to measure how many coins are in long-term savings, as opposed to being actively circulated?
And in the case of deleted wallets? What will those coins be considered?
What would be the best way to measure how many coins are in long-term savings, as opposed to being actively circulated?
And in the case of deleted wallets? What will those coins be considered?
In Quantitative Analysis of the Full Bitcoin Transaction Graph, Dorit Ron and Adi Shamir have analyzed statistics of ownership and circulation of bitcoins. One of the metrics they have investigated is the number of bitcoins that lie dormant in long-term savings rather than being actively used in trade.
They arrived at a figure of 78% of all coins being in savings account, where they define as savings any address which does not have any outgoing transaction. The problems with this metric are:
1. If addresses are rapidly generated and used up, circulated coins can exist in addresses matching this criterion. In fact, if everyone follows the guideline of never reusing an address, 100% of coins at all times will be in addresses which have never sent, regardless of how many are in circulation.
2. A savings address may have outgoing transaction; for example, if the address is emptied occasionally for a large purchase and then replenished over time.
3. It cannot distinguish between actual long-term savings which have not yet been liquidated and lost coins.
4. If a shared wallet is used, coins held in place by the wallet can represent circulating coins by clients, and vice versa. The first case is even more pronounced if payments are settled internally without the blockchain.
They also had this to say:
They arrived at a figure of 78% of all coins being in savings account, where they define as savings any address which does not have any outgoing transaction. The problems with this metric are:
1. If addresses are rapidly generated and used up, circulated coins can exist in addresses matching this criterion. In fact, if everyone follows the guideline of never reusing an address, 100% of coins at all times will be in addresses which have never sent, regardless of how many are in circulation.
2. A savings address may have outgoing transaction; for example, if the address is emptied occasionally for a large purchase and then replenished over time.
3. It cannot distinguish between actual long-term savings which have not yet been liquidated and lost coins.
4. If a shared wallet is used, coins held in place by the wallet can represent circulating coins by clients, and vice versa. The first case is even more pronounced if payments are settled internally without the blockchain.
They also had this to say:
Quote from: Adi Shamir
2. The question of how many bitcoins are dormant. Even though this was only one of our many findings, it was the one statistic that caught the attention of most commentators and web journalists. Webelieve that measuring this aspect is extremely important, since it implies some fundamental truths about whether the bitcoin scheme is used mostly to trade in goods or to speculate, and how thescheme will behave if its current users will change their current pattern of behavior. Once again , we have several possible approaches, and I would like to recruit your help in initiating a frank discussionin your community about how to measure this fundamental statistics in the most accurate way. We have spent a lot of time and effort in collecting all the data, but once we have it, it will be easy for us toadapt our measuring methodology based on your suggestions as a service to your community and as a contribution to the scientific study of the bitcoin scheme. However, there is one thing I would like toclarify: I do not want to have a target-shopping experiment, in which once you will see the results you will decide whether you like it or would prefer to change the methodology again in order to geta politically more appealing result. We thus propose to wait until some consensus emerges, and only then run the selected methodology on the data and announce the results.
2.1 In order to start this discussion, I would like to make an initial suggestion: Fortunately, the issue of dormant bitcoins can be made independent of the issue of common ownership of addresses(even though several comments we received argues the politically convenient point that since our decisions about the later point did not absolutely match the ground truth, our results about dormantbitcoins were completely invalid). Let us thus work entirely at the level of addresses, paying no attention to our attempt to find such common ownerships. We propose to compile the followingstatistics: For every amount of time x (e.g., in units of full months), we will measure how many bitcoins were deposited into addresses in transactions that took place more than x months before ourcutoff date (May 13-th 2012), and were not followed by any later transaction in which the address was one of the sending addresses.
2.2 Notice that if someone keeps moving his bitcoins every few days from one address to the next (which seems to be a relatively common behavior), we will not count these bitcoins at all as dormant(and in this sense we will underestimate the number of dormant coins) unless that user stopped doing this sometime before the relevant point in time (x months before the cutoff date), and in this casewe will only count these bitcoins once, in the last transfer which he executed (since all the previous transactions had a later outgoing transaction).
2.3. To deal with the issue of coins which were minted at the earliest period, when they had little value and could have been abandoned by early experimenters, we can also consider only transactionthat took place after a certain date.
2.4. To be fair, we will have to take into account that if we consider a large value of x, there were fewer coins at that time, so when we compute percentages, we will have to use in the denominatorthe number of coins that had been minted up to x months before the cutoff date.
What would be the best way to measure how many coins are in long-term savings, as opposed to being actively circulated? 2.1 In order to start this discussion, I would like to make an initial suggestion: Fortunately, the issue of dormant bitcoins can be made independent of the issue of common ownership of addresses(even though several comments we received argues the politically convenient point that since our decisions about the later point did not absolutely match the ground truth, our results about dormantbitcoins were completely invalid). Let us thus work entirely at the level of addresses, paying no attention to our attempt to find such common ownerships. We propose to compile the followingstatistics: For every amount of time x (e.g., in units of full months), we will measure how many bitcoins were deposited into addresses in transactions that took place more than x months before ourcutoff date (May 13-th 2012), and were not followed by any later transaction in which the address was one of the sending addresses.
2.2 Notice that if someone keeps moving his bitcoins every few days from one address to the next (which seems to be a relatively common behavior), we will not count these bitcoins at all as dormant(and in this sense we will underestimate the number of dormant coins) unless that user stopped doing this sometime before the relevant point in time (x months before the cutoff date), and in this casewe will only count these bitcoins once, in the last transfer which he executed (since all the previous transactions had a later outgoing transaction).
2.3. To deal with the issue of coins which were minted at the earliest period, when they had little value and could have been abandoned by early experimenters, we can also consider only transactionthat took place after a certain date.
2.4. To be fair, we will have to take into account that if we consider a large value of x, there were fewer coins at that time, so when we compute percentages, we will have to use in the denominatorthe number of coins that had been minted up to x months before the cutoff date.
Jump to: