BETI: Bitcoin Exponential Trend Index and technical analysis

[jamesc760]

Excellent work, OP!

I am currently bearish but having another look at this thread is making me re-think my outlook.

Thank you!

PS: BTC is really cheap right now. Note to myself: buy some coins today.

[1714471851]

1714471851

[1714471851]

1714471851

[1714471851]

1714471851

[1714471851]

1714471851

[1714471851]

1714471851

[1714471851]

1714471851

[jl2012]

Update:

Date:    11-Apr-2014
VWAP:    392.23
x:    1364
a:    0.00605
b:    -1.82543
Rsq:    0.88479
The day's expected price:    615.61
Predicted date for today's price:    26-Jan-2014
Days ahead:    -74.54
Daily price rank:    149
Predicted date for ATH ($1126):    25-Jul-2014
   
(See OP for explanation)   
   
   
   
https://www.wolframalpha.com/input/?i=e+%5E+%28+0.00604694820926662++%28+number+of+days+since+jul+17%2C+2010+%2Fdays+%29+-1.82542881007982+%29   

[jl2012]

On 11 Apr 2014, the price/trend ratio was the lowest since Feb 2013 (y-axis is ln(price/trend) )

[rpietila]

every single other poster in this forum I've seen who uploads his variant of a log linear regression chart (including well known Finnish investors clowns) post their pretty picture with a straight line, saying proudly "look, how close price is at all times to that line!", only showing how pathetic their understanding of TA is. The point, as exemplified by your second graph, is that you need to look at the *running* calculation of a regression, based on the data *available at the time*. And then it becomes clear that, at least in 2011, price stayed for a long time deep below the trendline predicted by regression.

Your rant is based on a wrong understanding of my model. It has always featured a monthly recalculation of the trend (which for practical purposes does not differ from a running calculation).

Also as you can see from the same link, my model which has the price data from 1/2009 (unlike all the models that arbitrarily ignore the data prior to Mt.Gox, including the one in OP) does very well with the 2011 slump, giving a buy signal at $2.28 (similar signal was given at $71 last summer and at $460 in 2014-3-31).

And, as opposed to the aforementioned Finnish clown, whether that happens (bull or bear market in the meantime) is *not* answered by regression.

Either I have subconsciously changed my mind and started to support you, or you have been misunderstanding my stance all along.

[rudius]

A little harsh, maybe?

To his defense, that Finnish clown does reference %'s below and above the trendline. Although, I agree that his assumptions that we won't stay below it for a long time because - well, just because - are his hopes, not real analysis.

I wasn t here in 2011 so you sure could help. But i would say that the landscape was quite different. Now we have a high rate of merchant adoption and every media is speaking about bitcoin. Things are going too fast. Of course, i will not take it for certain, but unless someone has a better argument than "it has happened in the past", i agree with him.

[ArticMine]

With the same dataset the current regression should correspond to rpietila's model. The main advantage I can see here is that one can easily tell what the prediction would have been at points in the past without clouding the issue with data points in the future. This is actually very useful. As for criticisms those that I made with respect to rpietila's model also apply here. https://bitcointalk.org/index.php?topic=400235.msg6192769#msg6192769

[jl2012]

every single other poster in this forum I've seen who uploads his variant of a log linear regression chart (including well known Finnish investors clowns) post their pretty picture with a straight line, saying proudly "look, how close price is at all times to that line!", only showing how pathetic their understanding of TA is. The point, as exemplified by your second graph, is that you need to look at the *running* calculation of a regression, based on the data *available at the time*. And then it becomes clear that, at least in 2011, price stayed for a long time deep below the trendline predicted by regression.

Your rant is based on a wrong understanding of my model. It has always featured a monthly recalculation of the trend (which for practical purposes does not differ from a running calculation).

Also as you can see from the same link, my model which has the price data from 1/2009 (unlike all the models that arbitrarily ignore the data prior to Mt.Gox, including the one in OP) does very well with the 2011 slump, giving a buy signal at $2.28 (similar signal was given at $71 last summer and at $460 in 2014-3-31).

And, as opposed to the aforementioned Finnish clown, whether that happens (bull or bear market in the meantime) is *not* answered by regression.

Either I have subconsciously changed my mind and started to support you, or you have been misunderstanding my stance all along.

If there were reliable valuation of bitcoin before MtGox, I'm happy to include that in my model. Anyway, it doesn't make sense to include Jan 2009 as the price was 0 and (log 0) is undefined

[rpietila]

If there were reliable valuation of bitcoin before MtGox, I'm happy to include that in my model. Anyway, it doesn't make sense to include Jan 2009 as the price was 0 and (log 0) is undefined

There is no reliable valuation, I have been searching for it, but the trades were so small and far between. In Jan 2009 the only trade I know of was conducted at $0.001. Later that year there are trades at $0.003 and $0.0054 at least.

 This dataset apparently shows that in 1-6/2010 the price meandered between $0.0028933 and $0.0075.

It is general knowledge that the opening of Mt.Gox raised the price significantly (to $0.05-$0.08). I have been saying for long that before 2010-7-7 there was never a bitcoin trade conducted at higher than $0.01. Of course I cannot prove it, but anyone can disprove it is he has the knowledge.

My trendline has been using the fixed price of $0.05 for all the months up to 6/2010. I have checked that it makes very little difference to alter the data according to individual trades between $0.003-$0.0075, and in my opinion it is not even honest research to do so, since there is not enough evidence to indicate that these isolated trades would be representative of a general price level, which Mt.Gox prices certainly are.

[oda.krell]

Your rant is based on a wrong understanding of my model. It has always featured a monthly recalculation of the trend (which for practical purposes does not differ from a running calculation).

Also as you can see from the same link, my model which has the price data from 1/2009 (unlike all the models that arbitrarily ignore the data prior to Mt.Gox, including the one in OP) does very well with the 2011 slump, giving a buy signal at $2.28 (similar signal was given at $71 last summer and at $460 in 2014-3-31).

Let me get this right... You have, by hand, *estimated* the pre-mtgox prices, and then boast that the regression analysis of those made-up prices gives you excellent results for the 2011 bubble?

You've got to be kidding, right? That's not how statistical modeling works.

In case you seriously need an answer to this: if you don't have data for some period, and you have no way to systematically extrapolate data for that period, you don't model that period. The End.

And about the glaring difference between your model and that of jl2012: you're correct that you recalculate the regression as well each month, but you don't *present* it that way. You keep posting the latest regression trend line, and then the discussion is always about that one, and how accurate it is, completely forgetting all the preceding ones.

jl2012's presentation on the other hand is much more "honest" if you want: it shows, in one graph, exactly what his model predicted at time step n, and what the actual price was at time n.

That way, everyone can see, at a glance, where a loglinear regression model was spot on, and where it missed the mark.

Which brings me to the second point...

Either I have subconsciously changed my mind and started to support you, or you have been misunderstanding my stance all along.

This is just too easy... I only need to go your post history of yesterday or so to find the following:

[...]
- Trendline comparison: we are now at -0.367 log units. The trendline is at $993 and rising $7 per day, conclusion: rock bottom

In other words, because we are drastically below the trendline right now, in your words "rock bottom", now is a good time to buy.

What this claim completely ignores however, and which is beautifully visible in jl2012's presentation, is that around October 2011 we hit "rock bottom" as well, but in reality we had about another halving (or more) of price ahead of us, i.e. it would have been well ahead of time to buy back.

It all runs down to the same criticism that applies to all log-linear price based methods:

They are an interesting tool to get an idea of what order of magnitude of price to expect for a given time, but they are *way* off at times, and trading advice of the form "we can only go up now, because: rock bottom" is in no way justified by the accuracy of the model.

[rpietila]

Also as you can see from the same link, my model which has the price data from 1/2009 (unlike all the models that arbitrarily ignore the data prior to Mt.Gox, including the one in OP) does very well with the 2011 slump, giving a buy signal at $2.28 (similar signal was given at $71 last summer and at $460 in 2014-3-31).

Let me get this right... You have, by hand, *estimated* the pre-mtgox prices, and then boast that the regression analysis of those made-up prices gives you excellent results for the 2011 bubble?

No, it does not give any selling signals for the 2011 bubble, because at that time the data is not yet fitting to a steady exponential model (long time flat and then explosive growth). I could not have used this model at that time.

But after the bubble, the bolded part above would have shown the buy point. That is now fixed as the "buy signal" area, due to the reason that it was a correct buy point. Using this definition for a buy signal, correct signal for 7/2013 was provided, and I am pretty sure the current one is also correct.

Sell signal was calibrated in 4/2013 and applied first time in 11/2013.

In case you seriously need an answer to this: if you don't have data for some period, and you have no way to systematically extrapolate data for that period, you don't model that period. The End.

I respect your opinion and will leave this to the readers. (To your credit, you probably did not have the above post of mine when composing yours, because it explains what has been done and why.)

...
That way, everyone can see, at a glance, where a loglinear regression model was spot on, and where it missed the mark.

It is available in my TA thread now, previously I relied on the assumption that readers are sort of knowledgeable on the field.

- Trendline comparison: we are now at -0.367 log units. The trendline is at $993 and rising $7 per day, conclusion: rock bottom

In other words, because we are drastically below the trendline right now, in your words "rock bottom", now is a good time to buy.

Yes, because a "buy signal" has flashed for the first time in 9 months, and we are 10% below it.

What this claim completely ignores however, and which is beautifully visible in jl2012's presentation, is that around October 2011 we hit "rock bottom" as well, but in reality we had about another halving (or more) of price ahead of us, i.e. it would have been well ahead of time to buy back.

His model does not have fixed "buy" and "sell" signals, afaik. So it is impossible to evaluate its goodness from a trader's perspective. My model has both buy signal (-0.3 when reached first time after a bubble) and sell signal (+0.45 and then follow the market). It has given 1 confirmed excellent buy signal (7/2013) and one confirmed excellent sell signal (11/2013). I am not claiming anything more, but I also don't know of any trendline that actually has performed better in timing the bitcoin price. 

"we can only go up now, because: rock bottom" is in no way justified by the accuracy of the model.

Again you are entitled to your opinion, just don't forget that I bought a castle with the proceeds I made with following my trendline. Bullshit walks.

[oda.krell]

[snip]

His model does not have fixed "buy" and "sell" signals, afaik. So it is impossible to evaluate its goodness from a trader's perspective. My model has both buy signal (-0.3 when reached first time after a bubble) and sell signal (+0.45 and then follow the market). It has given 1 confirmed excellent buy signal (7/2013) and one confirmed excellent sell signal (11/2013). I am not claiming anything more, but I also don't know of any trendline that actually has performed better in timing the bitcoin price.  

Agreed on that point. I disagree with a lot of your analysis, or more precisely: with the certainty with which you tend to present it to the newcomers, but you do make testable predictions. That's a big plus.

The castle... yeah. Good on you for that. But that's entirely unrelated to the question at hand. There are people on this forum that could buy 20 of those (by being early miners or investors), but they don't make market predictions. Keep those two items separate, please.

[Odalv]

Also as you can see from the same link, my model which has the price data from 1/2009 (unlike all the models that arbitrarily ignore the data prior to Mt.Gox, including the one in OP) does very well with the 2011 slump, giving a buy signal at $2.28 (similar signal was given at $71 last summer and at $460 in 2014-3-31).

Let me get this right... You have, by hand, *estimated* the pre-mtgox prices, and then boast that the regression analysis of those made-up prices gives you excellent results for the 2011 bubble?

No, it does not give any selling signals for the 2011 bubble, because at that time the data is not yet fitting to a steady exponential model (long time flat and then explosive growth). I could not have used this model at that time.

But after the bubble, the bolded part above would have shown the buy point. That is now fixed as the "buy signal" area, due to the reason that it was a correct buy point. Using this definition for a buy signal, correct signal for 7/2013 was provided, and I am pretty sure the current one is also correct.

Sell signal was calibrated in 4/2013 and applied first time in 11/2013.

In case you seriously need an answer to this: if you don't have data for some period, and you have no way to systematically extrapolate data for that period, you don't model that period. The End.

I respect your opinion and will leave this to the readers. (To your credit, you probably did not have the above post of mine when composing yours, because it explains what has been done and why.)

...
That way, everyone can see, at a glance, where a loglinear regression model was spot on, and where it missed the mark.

It is available in my TA thread now, previously I relied on the assumption that readers are sort of knowledgeable on the field.

- Trendline comparison: we are now at -0.367 log units. The trendline is at $993 and rising $7 per day, conclusion: rock bottom

In other words, because we are drastically below the trendline right now, in your words "rock bottom", now is a good time to buy.

Yes, because a "buy signal" has flashed for the first time in 9 months, and we are 10% below it.

What this claim completely ignores however, and which is beautifully visible in jl2012's presentation, is that around October 2011 we hit "rock bottom" as well, but in reality we had about another halving (or more) of price ahead of us, i.e. it would have been well ahead of time to buy back.

His model does not have fixed "buy" and "sell" signals, afaik. So it is impossible to evaluate its goodness from a trader's perspective. My model has both buy signal (-0.3 when reached first time after a bubble) and sell signal (+0.45 and then follow the market). It has given 1 confirmed excellent buy signal (7/2013) and one confirmed excellent sell signal (11/2013). I am not claiming anything more, but I also don't know of any trendline that actually has performed better in timing the bitcoin price. 

"we can only go up now, because: rock bottom" is in no way justified by the accuracy of the model.

Again you are entitled to your opinion, just don't forget that I bought a castle with the proceeds I made with following my trendline. Bullshit walks.

When the ruin is a castle then your agenda is pretty clear ...

[ArticMine]

If there were reliable valuation of bitcoin before MtGox, I'm happy to include that in my model. Anyway, it doesn't make sense to include Jan 2009 as the price was 0 and (log 0) is undefined

There is no reliable valuation, I have been searching for it, but the trades were so small and far between. In Jan 2009 the only trade I know of was conducted at $0.001. Later that year there are trades at $0.003 and $0.0054 at least.

 This dataset apparently shows that in 1-6/2010 the price meandered between $0.0028933 and $0.0075.

It is general knowledge that the opening of Mt.Gox raised the price significantly (to $0.05-$0.08). I have been saying for long that before 2010-7-7 there was never a bitcoin trade conducted at higher than $0.01. Of course I cannot prove it, but anyone can disprove it is he has the knowledge.

My trendline has been using the fixed price of $0.05 for all the months up to 6/2010. I have checked that it makes very little difference to alter the data according to individual trades between $0.003-$0.0075, and in my opinion it is not even honest research to do so, since there is not enough evidence to indicate that these isolated trades would be representative of a general price level, which Mt.Gox prices certainly are.

If one goes to the right side of the page and closes the graphs one finds daily data for both New Liberty Standard and Bitcoin Market in terms of 1 gram of gold via pecunix from January 16, 2010 to June 07, 2010.

[ArticMine]

...
Again you are entitled to your opinion, just don't forget that I bought a castle with the proceeds I made with following my trendline. Bullshit walks.

Actually the really tricky part was avoiding a million dollar plus "goxing" of the proceeds at the eleventh hour of the MTGox collapse.

[tuneman1980]

Great work jl2012.  I'm not trying to highjack your thread because I can't replicate your great analysis and calculations, but my website seems to go hand in hand with what you're doing here.  Please check it out and I hope someone else on this thread finds it useful as well. www.cryptocoinstats.com
 jl2012, I'll continue to read your updates.  Thanks again.

[rpietila]

...
Again you are entitled to your opinion, just don't forget that I bought a castle with the proceeds I made with following my trendline. Bullshit walks.

Actually the really tricky part was avoiding a million dollar plus "goxing" of the proceeds at the eleventh hour of the MTGox collapse.

"Avoiding" is the wrong word here, I just sold them in the open market to guys whose perception of the Mt.Gox risk was different enough from mine. I still need to find a way to explain the loss to taxman. And I still lost $250k. Of course now I tap myself on the back for being smart, but if Mt.Gox has continued operations and resumed withdrawals, I would not be so smug 

I don't know what problem oda.krell has with me. If someone can buy 20 castles, it means that he has invested better than me. How about giving us mortals some tips then? Oda.krell himself can hardly buy 3 used Ferraris... (this is a real-life example how one bitcoiner's wealth was described in a meetup - ability to buy 3 Ferraris)

Surely there is no guarantee that the "my" trendline has predictive ability. Actually it's enough to be correct 52% of the time. I believe there is more than 52% chance that the current buy signal is correct and no material downside is left.

Sorry OP for threadjack. I think you are doing good job.

[oda.krell]

I don't know what problem oda.krell has with me. If someone can buy 20 castles, it means that he has invested better than me. How about giving us mortals some tips then? Oda.krell himself can hardly buy 3 used Ferraris... (this is a real-life example how one bitcoiner's wealth was described in a meetup - ability to buy 3 Ferraris)

Don't worry too much about how my investments turned out. I can buy about as many used Ferraris as you have hair on your head :)

Seriously though, in case you didn't get the point I made: there are several people on btctalk compared to whom you are small fish. They still have the good sense no to walk around pretending to be knowledgeable TAers. Ergo: if they don't, you definitely shouldn't.

[rpietila]

I don't know what problem oda.krell has with me. If someone can buy 20 castles, it means that he has invested better than me. How about giving us mortals some tips then? Oda.krell himself can hardly buy 3 used Ferraris... (this is a real-life example how one bitcoiner's wealth was described in a meetup - ability to buy 3 Ferraris)

Don't worry too much about how my investments turned out. I can buy about as many used Ferraris as you have hair on your head

Seriously though, in case you didn't get the point I made: there are several people on btctalk compared to whom you are small fish. They still have the good sense no to walk around pretending to be knowledgeable TAers. Ergo: if they don't, you definitely shouldn't.

Asserting something that goes blatantly against provable and proven facts is considered lying, is it not, Mr. President?

(I am very much aware that most of the dozens of people who have more coins than me read the forum, but also know that my trading skills are above average evidenced by the results.)

[piramida]

I don't know what problem oda.krell has with me. If someone can buy 20 castles, it means that he has invested better than me. How about giving us mortals some tips then? Oda.krell himself can hardly buy 3 used Ferraris... (this is a real-life example how one bitcoiner's wealth was described in a meetup - ability to buy 3 Ferraris)

Don't worry too much about how my investments turned out. I can buy about as many used Ferraris as you have hair on your head

Seriously though, in case you didn't get the point I made: there are several people on btctalk compared to whom you are small fish. They still have the good sense no to walk around pretending to be knowledgeable TAers. Ergo: if they don't, you definitely shouldn't.

you guys with your petty dispute are turning one of the (few) threads with content on this subforum to another dick-measuring contest, sad and not interesting to read. just saying.

[rpietila]

You tell if either of us has brought any on-topic input to the thread! The other has to leave (or both, in case both are just blathering).