Bitcoin Forum

I've been studying the blockchain for a bit and I'm thinking about developing a metric to label bitcoins that are likely dead (ie the private key is no longer accessible). Do you guys have any advice for possible starting points?

How can you know whether a coin is lost or kept for the long term?

Well. If not used for a long time, It's dead. The only difference is it can be revived. However it does not have any influence in the market until it starts moving.
So... yeah

That's why it has to be a probabilistic measure based upon historical transaction rates combined with the time in storage. It is impossible to assert a deterministic difference between a bitcoin in cold storage and a bitcoin that is dead.

And yes there is an influence on the market once a coin has become dead, then supply is reduced. For a single transaction this isn't an issue, but we need to measure the rate at which this happens.

A wallet can change hands without moving, private keys are being traded, leave em alone!

We need to understand the impact coins that are inaccessible to the network will have on deflation

Why? I certainly have no need to know.

If you're interested, by all means, carry on with trying to find a suitable metric (although I wish you luck on that...) but just know that any proposal that has someone try to "resurrect" (i.e., steal) my stored coins, or anyone else's, will probably end poorly.

if you are trying to find "impossible" addresses, you could look for patterns and stuff that looks like words(ie, non-random).

Because I'm a mathematician and I enjoy economics. If the Bitcoin is to be studied academically and effect change in global monetary policy, then we need to understand fundamental things such as the rate of coin loss and better metrics for transaction volume. This is why bitcoin days destroyed was developed and why I'm designing a death metric for coins.

A recycling scheme would require a hard fork. We both know that isn't going to happen. Especially considering cold storage is popular.

Bitcoin addresses do have a verification scheme built in that removes nearly all impossible patterns (1 in 4 billion change of failure). However, I am looking for addresses that were once legitimate and have received money, but the private key has been lost

yes but if you remove the checksum, you could still put in random text. but the bitcoin would not be spendable afterwards.


in the other case, i would look at the address that are likely to be in the same wallet and see if they are active, other than that... there are only the option of seeing for how long the address have not been used.

The question is can we build a probability distribution that asserts within a confidence interval the chance a coin is dead. It should approach 100% after some set of iterations

Also cold storage coins would also suffer the same fate. It's not suppose to be a perfect metric. It is giving an effective bound on deflation from money taken out of normal circulation

Nice round numbers likely aren't lost, so you probably want to have two different rules.  For example, a 7.23512343 coin that hasn't moved in years might be a lost coin, whereas a 250 BTC coin that hasn't moved over that same period of time is much more likely to be the address for a private key printed and stored in someone's fireproof safe.   Also, the value at the time it was sent will matter as well.  I would be more likely to understand that someone would have abandoned a wallet with 4.0 BTC in it in late 2011 due to having forgotten the exact encryption passphrase used, whereas today with a 4.0 BTC wallet it would be worth it to try a cracking utility to try and regain access to the wallet.

Who would accept a private key from someone knowing that the original ownwer could have kept a copy. I understand the concept between family members etc but besides that I don't see any advantages.

Take a look at Casascius coins and bars.

In reality you always have to trust the party you buy a Bitcoin from if you want anonymity. If the guy you buy them from earned them on SR, they will come kicking your door in too, when they are looking into what the seller spend his coins on!

An untouched wallet or cascasious, where only mined coins are put in, is totally disassociated from any identity. In reality, if you spend it, it is no longer anonymous as if the guy you pay it to has his wallet linked to a bank account, someone can ask him where he got it from.

Thats the price of mainstream!

If we did develop an effective metric, then it is likely we can include a voluntary cold storage database. I also like the idea of noting certain integer amounts like 10 or 250 are likely large term savings. I think we could create a normalized scale between 0 and 1 and assign the higher values preference in a probabilistic measure in relation to time. Thus if a coin comes from a known cold store address, then its chance of being labeled as dead is very low regardless of time. If it is some non-integer real value that has been transacted several times and suddenly stopped moving for a long duration, then the probability of being dead increases considerably.

Then we also need to know the amount of fiat cash that is lost or accidentally destroyed each year, as well as the amount of gold that sinks to the bottom of the ocean and which could possibly be recovered.

:)

This isn't a fair assessment because in the first case fiat cash is highly inflationary. The rate at which they print money will always be great than money lost. In terms of gold, there is enough in the world supply that a few events like sinking ships will not affect the global price. We also do not know the total future supply of gold.

Economists do keep track of monetary hoarding. And people do measure supply of commodities. Why shouldn't we do this for the bitcoin?

Use Bitcoin Days Destroyed to determine hoarding volume. One of my favorite views is a running 30-day average of coin not spent in a minimum of a year (https://blockchain.info/charts/bitcoin-days-destroyed-min-year?timespan=1year&showDataPoints=false&daysAverageString=30&show_header=true&scale=0&address=).

This is like a measure of inverse hoarding, so the higher it goes, the less people are hoarding. It seems to have started peaking around the time Avalons came out and the price started spiking, which makes sense, old miners are upgrading and previous hoarders appear to be selling for fiat at a reasonable profit right now.

I've considered that BDD was the best starting point. I'd like to invent a metric to label addresses as probabilistically dead. Then we can measure the average yearly coin death rate. 

Okay, I can see how it might be a useful metric.

Could one not just as easily argue the reverse, especially as it pertains to untouched block rewards of the solo-mining era?

It seems probable that "death rate" is substantially reduced today from the days when bitcoins had little value at all. Any estimate would almost certainly need to consider different "eras", applying disparate rules to each.

As a first step, I would think it worthwhile to look at some statistics for mean-time-between-spends as a function of input size, possibly splitting up the data in "eras".

The mean-time-between-spends would be easy enough to calculate (I gather from other threads that you are looking at blockchain parsing) and might help in picking out types of inputs (size, past behavior, etc) that are likely to be "dead".

The one point worth making here is that there is no continuity from really long m-t-b-s to "dead", so the temptation to use it as an indicator of a similar input's likelihood of being dead should probably be avoided (and is not the suggestion I'm making).

It is important to note that the notion of death is probabilistic and one can never label a coin dead. It is only useful to discuss the idea that a coin is unlikely to ever be spent and thus is effectively out of circulation. I like your idea of eras.

And yes I am looking into blockchain parsing. I can write some software with R to conduct an interesting anaylsis. 

I realize that. My point was simply that, while it may be tempting to conclude that as mtbs increases, so does the probability of the input being dead, that may not be the case. Specifically, I wouldn't necessarily expect Pdeath to be a monotonic function of mtbs (or "dormancy", if you will).


There are a couple of nice parsing code bases available, as was mentioned on the thread where you requested that. I'd recommend Znort's (available as a github repository). A really easy place to start is the following post in his 410 richest addresses thread:

https://bitcointalk.org/index.php?topic=92423.msg1596690#msg1596690

If you're on Linux it will literally take you 5 lines in a terminal to get going--and that includes one for installing build tools ;-).