.

[molecular]

A lesson that I learnt the hard way: you should never dump all.

Lessing I learned the hard way: you should never go all-in.

[1714991017]

1714991017

[1714991017]

1714991017

[1714991017]

1714991017

[Nemo1024]

A lesson that I learnt the hard way: you should never dump all.

Lessing I learned the hard way: you should never go all-in.

Yeah, and that too  

I am now trying to keep a balanced holding of BTC and $, with a slight bias towards one or the other, depending on the prevailing trend.

[powpow]

A lesson that I learnt the hard way: you should never dump all.

Risk to hold outweighed risk of selling.

Assume I sold 250 BTC @ 138

If the price goes up to $170 I will probably buy back in and still have the same amount of $.

250 * 138 = $34,500
$34,500 / $170 = 202 BTC (same USD value)

At this point, I feel the risk is higher that the price is going to drop significantly.  I think I'll be buying in somewhere around ~100, which will give me a gain of 100 BTC.

There is no risk for selling now, but there is a risk for holding. Selling everything is always a safe bet.

[Nemo1024]

There is no risk for selling now, but there is a risk for holding. Selling everything is always a safe bet.

It wasn't for the poor buggers who sold everything @ $30 before the $240 spike. Not that I am saying we are in the same situation now, but you never know...

[powpow]

There is no risk for selling now, but there is a risk for holding. Selling everything is always a safe bet.

It wasn't for the poor buggers who sold everything @ $30 before the $240 spike. Not that I am saying we are in the same situation now, but you never know...

Still a safe bet, any profit is a good thing.  It's when you start stretching when you have a real problem.

[adamstgBit]

There is no risk for selling now, but there is a risk for holding. Selling everything is always a safe bet.

the risk you perceive for holding bitcoin is greatly exaggerated, and you fail to recognize the risk with holding everything in USD

[powpow]

There is no risk for selling now, but there is a risk for holding. Selling everything is always a safe bet.

the risk you perceive for holding bitcoin is greatly exaggerated, and you fail to recognize the risk with holding everything in USD

lol

[adamstgBit]

There is no risk for selling now, but there is a risk for holding. Selling everything is always a safe bet.

the risk you perceive for holding bitcoin is greatly exaggerated, and you fail to recognize the risk with holding everything in USD

lol

 

its true dude

bitcoin is happening now regardless of risk

USD is going down regardless of economic forecasts.

cash is king until its not.

bitcoin is gold! gold i tell you!

[RationalSpeculator]

Do I understand correctly that you expect the price to go to around $60 before the end of the year?

How much percent chance do you give this? 50%? 90%?

Also, I'm confused on what you base this?

That is the median case, but this is just a toy.

What I am looking at is an indicator which shows the likelihood that an intermediate reversal has occurred. It works by comparing how BTCOBV and $OBV track compared to price development in order to detect what regime of feedback effects is dominant.

Then I go back to look at the distribution of the change in price from bin to bin in the last leg before the last reversal and take the compliment of it to form a random variable going forward, there are actually 5 random variables involved which have some dependency re. the development of the reversal indicator, to provide a random trial of where it might turn again in the future, this is the output of the trial.

I am not actually certain that it is meaningful to assign a point probability to a particular outcome. It would be more meaningful to say, for instance, that there is a percentile case for price being above or below a target, or before or after a target. The median case as a rule of thumb would be the 50th percentile presuming a well-behaved Gaussian distribution. But I don't think that is a safe assumption. For one thing, I know that there are dependencies that I don't have worked into the (Markov chain) Monte Carlo yet.  

That is chinese for me. I have no idea what your answer to my question is.

I'm willing to pay to get your opinion if that is what it takes.

In such case, I would ask: explain like I'm a 5 year old.

Are you able/willing to?

I'm not sure that I can but let's start from the beginning.

First of all, I don't really use time at all in my conception of this process we call the market. Without understanding this we are not speaking the same language. The very first thing I do before anything else happens is to compress the raw time series into bins of similar sized Bitcoin volume. The size of these bins is not arbitrary, but is based on a limit having to do with an extreme of price expectation as defined by ATH peaks. And it attempts to track how this limit slowly evolves over time, which means that the mean size of the bins grows slightly smaller as we progress.

It is not possible for now to explain exactly why I do this but the thing to understand is that this is my starting point, all further analysis is built from this foundation. I will tell you that the minimum bin size is currently about 5,000 BTC, so each discrete chunk of data in the analysis is at least that size. For the purpose of the analysis I have been describing in this thread it is sufficient to consider that I am working from chunks of data (High, Low, VWA, BTCVol, $Vol, Timestamp) that have a minimum size of about 5,000 BTC.

You have probably observed that OBV generally tracks up and down with price, but not exactly, proportionately. That is, when price goes up, OBV also generally goes up, but not always by the exact same proportion. BTCOBV and $OBV also track this progression of price differently. In order to see how this difference in how OBV tracks price changes with the progress of the market I find it useful to look at a ratio of OBV to price. Except that if you were to do this for each and every bin (discrete sample) the result would be very noisy and it would be difficult to see a clear trend. In order to get around this problem I will look at how the median value, of say the las 100 samples, of OBV compares with the median of the last 100 samples of price. Kind of like a moving average, except instead of using an average I use the median value. There is very good reason to use a median instead of an average as a median is less sensitive to "shocks", that is it does not jump as much if there is a single sample in your series that is much higher, or much lower than the rest of the series. And so, the result tends to be less 'noisy'.

I do this separately, for BTCOBV/Price and $OBV/Price, and then look at the difference between them. What happens is that as price increases above an equilibrium point the difference between these ratios increases, and as price declines below an equilibrium point the difference increases in the other direction. You can 'sharpen' the response of this difference by calibrating the number of samples you are looking at. That is, if you see that there is 115 samples between an intermediate High price and Low price it helps to set the length of your 'moving median' to the same length. Then it is easier to see when an intermediate reversal has actually happened.

The tricksy thing to know is how 'valid' this calibration is, that calibration is one of many assumptions that has to be made in order to attempt a projection going forward. The way I try to measure this, all things being relative, is to see how high is high, and how low is low compared to what has happened before. This BTCOBV/Price - $OBV/Price wiggle goes up and down, kind of like an oscillator, but it isn't 'normalized'. In order to normalize it I will look at what the Percent Rank of the current value is compared to a certain number of samples of the indicator going backward. That is, if going backward 100 samples there are just as many samples above the current value as below it then the Percent Rank is 50%. By doing this I am hoping to get a gauge of the statistical significance of how high, or how low the current value is compared to the past.

But here you have to make another one of those assumptions. Generally, I think the most recent data is more important, but by using too small a sample for comparison there is a risk of 'over fitting' the data. As a rule of thumb when I am trying to set the calibration for how far back to look for this statistical significance, if I think we are at a peak I will take that value and go back until I find a value that is at the same height. Then if it is a high peak this calibration is set to 100%. I generally find that point about twice a many samples back as the calibration used for the 'moving median' of the reversal indicator. This is important—because the sample 'bins' are of similar volume the assumption that is being made is that similar behavior can be inferred for a similar passage of BTC volume in the future. If, for instance, you observe that the behavior going forward is not, in fact, similar you need to go back and reassess your assumptions.

Now, for the price projection... Here I look at the distribution of the change in the VWA from bin to bin. That is, VWA for the current bin divided by the VWA for the last bin, etc. So if the VWA for the current bin is $141.80, and the last bin is $140.38 then that differential is .990. Given the assumption of similar behavior for similar passage of volume I will then generate a random variable based on the distribution of those differentials for a certain calibration interval. So, if the current price is $141.80 and the random variable returns .990 I multiply that by the current price to yield a random walk to a projected value for the next bin of $140.38. If I believe that price will go down, and I am taking samples from increasing prices in the past I will invert the distribution in order to obtain am approximate complementary distribution. So if a sample from that distribution is .99 I subtract it from 2 to get 1.01... And do this for every sample to get an approximate complement for the distribution. (Ack! No) So if a sample from that distribution is .99 I invert it (1/.99) to get 1.01... And do this for every sample to get an approximate complement for the distribution.

This is a part that I am still working on. In order to simulate the progression of the reversal indicator into the future I plug in values from the random variables and just calculate it out like I would as if I were using actual data. It requires price, BTCVol and $Vol. Price and BTCVol are for now treated as independent variables with no dependence, and $Vol is derived from price and BTCVol. I know there are probabilistic dependencies that I am not considering for this simulation, I just don't have them all worked in yet. That is why I say it is a toy, there is rigor that I know should be included that has not been included yet.

As well, I am treating the random variable for time as independent, which is really not part of the calculation at all, but the progression of time in the projection has a similar random distribution as a sample in the past. Same goes for probabilistic dependencies here that have not been included.

So the random trial progresses until there is an indication of a reversal by the calibrated reversal indicator and the price and the time coordonate are output, this is what is shown on the scatter plot. Normally I would do about 200 trials. This takes about three hours. If I am doing a smaller test I might just do 20 trials to see if everything is working. Some trials throw out a null error if no reversal is indicated, but I think this could be useful too. Right now I am considering that a large number of error outputs could indicate that we are moving int a regime where this particular methodology is not as useful, or it could indicate that some of the assumptions need to be revised.

It is very much a work in progress.

I'm really sorry to say your above explanation was not of help to me.

I just don't understand your language. 

Thank you though for having taken the time to try to explain it to me chodpaba.

[bucktotal]

interesting stuff as always. my analysis highlights $125 if we go that way soon too.

some Q's if u dont mind. can you try to describe where you see noise entering into your model. or maybe just describe what are the largest contributors to variance?

is it possible that you have either 1) two overlapping distributions, or 2) 1 primary distribution with some artifactual noise?

[bucktotal]

i understand that you bin in volume, not time. you have made that point quite clear 

by "inverting values" do you mean take the absolute value = | d$/dt  |  ?

and what if trade vol is not a constant over time? lets say that it is decreasing slowly at some rate. in that case, counting trade vol bins would cause bigger errors as "volume required for reversal" or "time" of projection is increased. i guess you see that as it starts getting really messy the farther out in time the projection is.

[bucktotal]

I actually do count the default minimum value for volume-per-bin as slowly decreasing. It is a line that is slowly decreasing based on a calibration at the $32 peak and a calibration at the $266 peak.

Yes, it does get messier the farther out you go. Dependencies that are easier to see in the actual data start to break down for the projected variables, and it starts looking less and less like 'natural' data. Going out as far as I am trying to go with this method is a real stretch.

do you think adding some more calibration points to the default min vol-per-bin line would affect things? the gox market has definitely changed a couple of times since $266, imo

[bucktotal]

cluster is getting tighter. nice. to remove the (obv?) noise, perhaps consider: a) a z-score filter, say 3 sd  or b) colormap each value with its z-score, or c) just cut it off at 8 weeks


 

[derpinheimer]

Can you help me understand

1. How can this chart [Or, the theory behind it] be dis-proven?

Say for example, Novermber 4th, the price  is $175. This is an outlier yet it is "correct"... however, that means  at least 30 other data points are wrong.

So, how do you make use of the chart? It seems like the only way it can be "right" is if the price is within the 2 red vertical bands, E.G. $75-90 Oct. 7 thru Oct 27. Outside of that? Not enough data.

2. If the band is the only area that matters, why chart outliers? If they are hit, it means the system is mostly saying, "Tomorrow the price will be between 100 and 200 USD." Which really isnt valuable information. I guarantee to you that it will. But does this really help you or me in our trading decisions? No.

I dont mean to come off as rude; I really like your analysis. Its just this chart is throwing me off.

[adamstgBit]

Can you help me understand

1. How can this chart [Or, the theory behind it] be dis-proven?

Say for example, Novermber 4th, the price  is $175. This is an outlier yet it is "correct"... however, that means  at least 30 other data points are wrong.

So, how do you make use of the chart? It seems like the only way it can be "right" is if the price is within the 2 red vertical bands, E.G. $75-90 Oct. 7 thru Oct 27. Outside of that? Not enough data.

2. If the band is the only area that matters, why chart outliers? If they are hit, it means the system is mostly saying, "Tomorrow the price will be between 100 and 200 USD." Which really isnt valuable information. I guarantee to you that it will. But does this really help you or me in our trading decisions? No.

I dont mean to come off as rude; I really like your analysis. Its just this chart is throwing me off.

the way i see it, this thread is an indicator, and its a pretty good one...

[bucktotal]

cluster is getting tighter. nice. to remove the (obv?) noise, perhaps consider: a) a z-score filter, say 3 sd  or b) colormap each value with its z-score, or c) just cut it off at 8 weeks

I guess I don't understand what you mean by z-score in this context... 

you have a set of a couple hundred data points and it looks fairly gaussian besides the outliers outside of the cluster. each value in the set has z-score = x-mean/sd.   there are a few ways you could clean up the data if you wanted, and one way would be to remove data with a z>~3 maybe. z-score colormap would just look nice    

[sturle]

Lol what?
Guess what will happen to USD when dollar will loose it's monopoly on oil.
Hint

Much of the worlds oil trade already happen in other currencies than USD.  Most of Europe, where oil is produced and consumed, trade in EUR.  The price is usually given in USD, but that's just for convenience.  Converting between EUR and USD is a trivial matter.

As long as USSA is the world's largest oil importer by far (2 times China), USD will be the predominant currency for trading oil.

[Patel]

In about 5 years oil is going to be useless because you can run all the worlds machine with a infinite supply of cosmic energy. Nikole Tesla patented his machine in 1901

Just wanted to add that in their with the oil talk

[notme]

In about 5 years oil is going to be useless because you can run all the worlds machine with a infinite supply of cosmic energy. Nikole Tesla patented his machine in 1901

Just wanted to add that in their with the oil talk

Cool story.  Could you link to the patent that describes how it works?  All US patents are public record.

[Nemo1024]

In about 5 years oil is going to be useless because you can run all the worlds machine with a infinite supply of cosmic energy. Nikole Tesla patented his machine in 1901

Just wanted to add that in their with the oil talk

Cool story.  Could you link to the patent that describes how it works?  All US patents are public record.

Must be this one: http://www.google.co.uk/patents/US685957

And another link: http://fuel-efficient-vehicles.org/tesla-flying-machine/Tesla-aether-dark-radiant-energy.php