Future price of bitcoin - logarithmic chart

[Zarathustra]

What, I don't even... Do you even math? How does that correlate to anything at all?

I guess, it correlates to a different R squared.

At least the OP's chart had an idea of where it wanted to go, your chart makes my frontal lobe hurt trying to understand what possessed you to put that line there.

I am seriously not trying to mock you or anything. I just don't follow what your point is.

I guess you are tying to say that bitcoin will fail?

No, I don't. I think, it will work until civilization will fail again.

Care to make a bet on it? Lets say that you owe me 1btc in 2020 if bitcoin is successful and I will owe you $420usd which is the ATH of bitcoin of today. Seems fair right? tollface.jpg

I'm not stupid. I made a similar bet this year and I won.

I mean bitcoin will fail so you would be making out like a bandit Your chart says that we will never break the ATH again!

You are wrong. It doesn't.

Upon coser examination it looks like your chart is still bullish, just a lot lower

Yes, and therefore it implicates that ATH will be broken later.

But seriously, what is up with this chart you made? I want to understand your position.

My position is, that the blue line depends on the R squared that we choose.

[CMMPro]

I not sure you understand what R^2 means.

We don't "choose" R^2, linear regression is used to find a formula (for the simple slope in this case) that describes the line as close to "1.0" as possible....a value of 1.0 is a perfect correlation between the line and the data. 1.0 is impossible unless all data points are on that slope. Since this isn't a straight line 1.0 isn't possible....so you calculate the best fit, the highest R^2 value we can find.  

I also understand the philosophers argument that you are stating....since the bitcoin market never closes...how can anyone determine the closing bell price in order to chart it?

Even without a closing bell we still have the ability to chart all trades....I don't think that is what is happening in this chart, but it could be done to make a more accurate linear regression determination.

In this chart I think he/she is using the daily indicator which is likely just the end of the day at midnight gmt.


Edit: http://en.wikipedia.org/wiki/Coefficient_of_determination

[Enky1974]

try again starting from november 2011 low at 2$, leaving out the first bubble of 2011 and compare results

[oda.krell]

Well if you are choosing point then you are just drawing a line where you want it to be.  It is a pretty line.  I like that it is blue.  The complete lack of antialiasing give is a kinda retro vibe. Not much value beyond that.
 

No, it's not me who chooses the data points. What, if the computer chooses only monthly (median) or yearly data points?

I'm not a mathematician, I am only a philosopher and always happy to learn something. That's why I asked the Genius. Is it mathematically impossible to paint different lines? Who chooses R squared?

Hehe, don't take DeathAndTaxes' snark too serious. 't was a funny one I have to say.

Anyway, I think I see your point/question: you basically wonder about the (arbitrarily) chosen time resolution. I.e. what result would the regression deliver if, instead of, say, hourly data, we daily, or instead of daily, weekly.

The answer is: a) the result might look slightly different, but not much. b) the finer grained your time resolution, the more accurate and possibly: predictive your analysis will be, because you have more data points.

In short: if you get a vastly different result, for example by using 'monthly closing price on btc-e' or some shit like that, you're cheating :D

[LOADING.READY.RUN]

...and some more things to consider: What happens when you introduce a weight function related to the trading volume that actually took place at that time/price? And should the weight be determined from the number of BTC traded or USD traded? Then of course you couldn't use a simple regression method any longer, but you'd have to move on to Chi-square or the likes.

Not saying that you have to do that, just take the results from a simple regression with a grain of salt. An analysis is always based on numerous assumptions. You should just be aware which your assumptions are.

[oda.krell]

...and some more things to consider: What happens when you introduce a weight function related to the trading volume that actually took place at that time/price? And should the weight be determined from the number of BTC traded or USD traded? Then of course you couldn't use a simple regression method any longer, but you'd have to move on to Chi-square or the likes.

Not saying that you have to do that, just take the results from a simple regression with a grain of salt. An analysis is always based on numerous assumptions. You should just be aware which your assumptions are.

Or, you know, you could still do linear regression, but on daily VWAP, once for $V and once for btcV and see how much they differ. Not different enough to really matter, I'd guess.

[Xer0]

i doubt any form of finite values can rise exponentially and thus infinite
that means the rich get richer and richer for ever and at some time normal people can not even afford a satoshi
deflationary tough, a currency can not deflate more than the economy behind

i think any sort of logarithmic function would aproximate better

[notme]

i doubt any form of finite values can rise exponentially and thus infinite
that means the rich get richer and richer for ever and at some time normal people can not even afford a satoshi
deflationary tough, a currency can not deflate more than the economy behind

i think any sort of logarithmic function would aproximate better

http://en.wikipedia.org/wiki/Sigmoid_function

Also @Zarathustra: Please do some reading, your comments don't make much sense to those of us who understand the math that went into the first chart. http://en.wikipedia.org/wiki/Least_squares http://en.wikipedia.org/wiki/Linear_regression

[pabloangello]

But lets say that one day every avarage people that doesn't feel good with using computers will be able (becouse of some new BTC tech) to have Bitcoin wallet and easily pay/recieve money with BTC like they learned to use plastic credit cards before. Maybe then such a graph is possible. What do you think?

[LOADING.READY.RUN]

i doubt any form of finite values can rise exponentially and thus infinite
that means the rich get richer and richer for ever and at some time normal people can not even afford a satoshi
deflationary tough, a currency can not deflate more than the economy behind

i think any sort of logarithmic function would aproximate better

I think for certain growth phases an exponential function can be the best approximation. However, I'm well aware that this holds only for a limited period of time. At some point growth will be limited by the natural boundaries of the system, and bitcoin will slow down its growth at maybe 1000, 10000 or even 1 million USD. Where that actually will happen, no one knows yet.

[Zarathustra]

i doubt any form of finite values can rise exponentially and thus infinite
that means the rich get richer and richer for ever and at some time normal people can not even afford a satoshi
deflationary tough, a currency can not deflate more than the economy behind

i think any sort of logarithmic function would aproximate better

http://en.wikipedia.org/wiki/Sigmoid_function

Also @Zarathustra: Please do some reading, your comments don't make much sense to those of us who understand the math that went into the first chart. http://en.wikipedia.org/wiki/Least_squares http://en.wikipedia.org/wiki/Linear_regression

I guess you math cracks are right, but I also guess that a linear regression over the whole lifespan of Bitcoin doesn't make sense.
With time passing, the line is flattening out.

https://bitcointalk.org/index.php?topic=322058.msg3469380#msg3469380

[notme]

i doubt any form of finite values can rise exponentially and thus infinite
that means the rich get richer and richer for ever and at some time normal people can not even afford a satoshi
deflationary tough, a currency can not deflate more than the economy behind

i think any sort of logarithmic function would aproximate better

http://en.wikipedia.org/wiki/Sigmoid_function

Also @Zarathustra: Please do some reading, your comments don't make much sense to those of us who understand the math that went into the first chart. http://en.wikipedia.org/wiki/Least_squares http://en.wikipedia.org/wiki/Linear_regression

I guess you math cracks are right, but I also guess that a linear regression over the whole lifespan of Bitcoin doesn't make sense.
With time passing, the line is flattening out.

https://bitcointalk.org/index.php?topic=322058.msg3469380#msg3469380

Right, but if you change the line by selecting a subset of the data, you should be able to describe your methodology and a reason why it should be accepted as a model.  I have a feeling your methodology was "this looks good".

[Zarathustra]

Right, but if you change the line by selecting a subset of the data, you should be able to describe your methodology and a reason why it should be accepted as a model.  I have a feeling your methodology was "this looks good".

Yes, pure guessing based on the assumption that the 8 month-period of October 10 to June 11, in which the price went up thousandfold, will never be repeated again.

[mmortal03]

Right, but if you change the line by selecting a subset of the data, you should be able to describe your methodology and a reason why it should be accepted as a model.  I have a feeling your methodology was "this looks good".

Yes, pure guessing based on the assumption that the 8 month-period of October 10 to June 11, in which the price went up thousandfold, will never be repeated again.

Then, instead of pure guessing, just start a logarithmic regression line after June 11 and post that. Just drawing an eyeballed line IS simply less legitimate.

[Zarathustra]

Right, but if you change the line by selecting a subset of the data, you should be able to describe your methodology and a reason why it should be accepted as a model.  I have a feeling your methodology was "this looks good".

Yes, pure guessing based on the assumption that the 8 month-period of October 10 to June 11, in which the price went up thousandfold, will never be repeated again.

Then, instead of pure guessing, just start a logarithmic regression line after June 11 and post that. Just drawing an eyeballed line IS simply less legitimate.

1) I'm completely free to paint what I want.
2) I think, linear regression doesn't make sense even if we exclude the first Bubble.

[oda.krell]

Flashback time! This entire discussion reminds me so much of an exchange I had several months ago on the Wall Observer thread with a biological statistician guy (whose name I forgot, but he had a caricature of Darwin as his avatar I think), who didn't seem to be clear about the assumptions he made in performing his price analysis.

Anyway, I think this discussion suffers from a lack of mutual understanding. Those who know how regression works reject Zarathustra's claim that one might just as well draw random lines on the chart. And they're right of course. Zarathustra on the other hand (correct me if I'm wrong) probably feels the regression analysis is somewhat arbitrary as well. And he's right as well.

Why?

Because you make (at least) the following two assumptions:

1) It is possible to extrapolate from historical data. That's kind of a big assumption, but on the other hand: what else would you do? The only thing we *can* work with is historical data.

2) If you apply linear regression analysis to the entire, unpartioned historical data on a log chart, you assume there is exactly one linear price function (which you then attempt to approximate).

And the latter assumption is much more contentious, in my opinion.

Please note that I'm not saying that linear regression in general is useless to predict bitcoin price, just that it is important to be clear about the assumptions you make. So if you post a chart with a single big line like the one greaterninja posted, then the single function assumption needs to be addressed, because I'm sure a lot of people would be more critical if they knew this assumption is necessary to get this result.

[Zangelbert Bingledack]

It's arguably the best starting assumption, though. At least a straight line that is a best fit for the data is the simplest starting assumption. Occam's razor suggests not making it more complicated without particular reason, and it seems to me there's enough shooting in the dark as is that there's a very high bar on adding more complexity to the trendline.

[oda.krell]

It's arguably the best starting assumption, though. At least a straight line that is a best fit for the data is the simplest starting assumption. Occam's razor suggests not making it more complicated without particular reason, and it seems to me there's enough shooting in the dark as is that there's a very high bar on adding more complexity to the trendline.

True. You'll note I said that I'm not rejecting linear regression at all. Just that I want everyone to be clear about the assumptions made.

[Zarathustra]

It's arguably the best starting assumption, though. At least a straight line that is a best fit for the data is the simplest starting assumption. Occam's razor suggests not making it more complicated without particular reason, and it seems to me there's enough shooting in the dark as is that there's a very high bar on adding more complexity to the trendline.

True. You'll note I said that I'm not rejecting linear regression at all. Just that I want everyone to be clear about the assumptions made.

For me, it is useless in this case, because a thousandfold price increase within some months is possible exclusively in the very beginning. Or does somebody really believe that this could be repeated later in the game?

[BitChick]

It's arguably the best starting assumption, though. At least a straight line that is a best fit for the data is the simplest starting assumption. Occam's razor suggests not making it more complicated without particular reason, and it seems to me there's enough shooting in the dark as is that there's a very high bar on adding more complexity to the trendline.

True. You'll note I said that I'm not rejecting linear regression at all. Just that I want everyone to be clear about the assumptions made.

For me, it is useless in this case, because a thousandfold price increase within some months is possible exclusively in the very beginning. Or does somebody really believe that this could be repeated later in the game?

Why not?  We have just scratched the surface of what Bitcoin can be.  We are still on the ground floor.  We have not even begun to see what will happen when Wall Street gets involved with the Winkelvii trust.  I have heard people say that anything under $1000 is cheap.  I am really starting to believe that.  Call me crazy but in a year I think that this will be proven.