:

[deepceleron]

I think difficulty will level off at the next recalculation, and it will probably take near 14 days to get there. Potential miners are already cancelling their video card orders or trying to return or restock their hardware bought in the past week. We've learned that the prices on exchanges don't have to always go higher. In the next month ATI won't sell more Radeon HD GPUs than have ever been made. It will take a big bump in the exchange rates for people to go gung ho in buying new mining hardware again.

It will take about 11 days to get there. And for it to level off after that the average price until then would have to be about $8.

Blocks 133056 -> 135072: 12 days 8.8 hours

New difficulty-equivalent hashrate from first 48 hours of this round: 1581572 -> 1563027, -1.2%

Current trading price: $14.25

Buy energy hit stiff sell order resistance at $16, and stagnating at that price was enough to trigger some more cashout. I don't see publicity driving more get-rich-quick money in, as markets have settled in since gox 2.0 and there is not much to report. Mining with new hardware is a break-even proposition. We are approaching the 'viable currency' phase of bitcoin's life.

[deepceleron]

I think difficulty will level off at the next recalculation, and it will probably take near 14 days to get there. Potential miners are already cancelling their video card orders or trying to return or restock their hardware bought in the past week. We've learned that the prices on exchanges don't have to always go higher. In the next month ATI won't sell more Radeon HD GPUs than have ever been made. It will take a big bump in the exchange rates for people to go gung ho in buying new mining hardware again.

It will take about 11 days to get there. And for it to level off after that the average price until then would have to be about $8.

Blocks 133056 -> 135072: 12 days 8.8 hours

New difficulty estimate from first 48 hours of this round: 1581572 -> 1563027, -1.2%

Current trading price: $14.25

Buy energy hit stiff sell order resistance at $16, and stagnating at that price was enough to trigger some more cashout. I don't see publicity driving more get-rich-quick money in, as markets have settled in since gox 2.0 and there is not much to report. Mining with new hardware is a break-even proposition. We are approaching the 'viable currency' phase of bitcoin's life.

Not too shabby. Do you have a call for 139104?

If difficulty can be constant for a while, miners won't see exponentially diminishing returns looming on the horizon, and may gain confidence in investing in hardware equity that won't turn a profit for three months. I see the difficulty growth barely on the positive side in the next two cycles, with about 50% profit-taking, 40% holding, and 10% reinvestment after, and new mining stronger in October when energy costs are free instead of double (heating your home in winter instead of making your air conditioning run in summer).

[sadpandatech]

nice nice. I see it brings it closer to my original guesstimations. Though I see you now show a drop for 139104. Care to elaborate?


  One other factor I am having a hard time with is the botnet factor. and no, not the pools trouble of last week for anyone else thinking that. Just that keeping track of the total amount they are adding into the net is a pain since they are much more erratic than the pools. Or, atleast they were until they decided again it was more profitable to find a better way to 'fit in' than it was to ddos everyone.

  I am sure you watch the network total hash like I do. I noticed a decline after the last retarget had normailized from its usual dip. It was down by about .8 - 1.2 Thash that I could not account for with normal fluctuations, checking the big pools were not temp out and casual miner dropouts. This was from about 12 hours after retarget up to about 16-18 hours ago, where at which time it grew from the back to normal rate to what it is now.

  I'm having a hard time justifying that with gains from new onlined hardware, etc. Obviously some percentage up is expected and we know even with the difficulty even small ops are still onlining. But, if I may risk putting a guess on it, it seems about 20% of the new growth is not user generated....  I know some of you are gonna see 20% and think holy crap, the zombies are gonna eat our brains but let me repeat I am only refering to 20% of the 800-1200Ghash growth or roughly 160-240Ghash worth that I suspect of currently being botnets....


edit; just wanted to point out that I say 'guestimations' because thats just what mine are. I am certainly not doing any where near the lvl of calculation you are to come to your numbers.

edit2; same with my nethash numbers. Though I feel confident about my bot figures they are derived from semi-casual observation and math done in my small brain.

[skyhigh]

I will disagree on both predictions by chodpaba. Next difficulty will have a very hard time to get as high as 1.79 mill. Difficulty after that will have very hard time to drop as much as 1.64 mill. I think a more natural growth of the network will take over and we will go higher at a slower pace.

I predict 1.7 mill next difficulty and 1.8 after that followed by close to 2.0 mill.

[skyhigh]

Sorry, you are right. I was thinking too much that your median prediction is more of a hard target.

[skyhigh]

I will disagree on both predictions by chodpaba. Next difficulty will have a very hard time to get as high as 1.79 mill. Difficulty after that will have very hard time to drop as much as 1.64 mill. I think a more natural growth of the network will take over and we will go higher at a slower pace.

I predict 1.7 mill next difficulty and 1.8 after that followed by close to 2.0 mill.

There is no actual disagreement.

The targets you are asserting are well within the mid-range of my projection.

I would like to assert that it is not as important for business planning purposes to precisely hit a particular target. The important thing is to accurately and reliably characterize the probability that a certain target will be realized. A business, or investor, that bets the farm on a single outcome will not stay solvent for long. The thing that is important to know is what the likelihood of certain conditions will be. If you put all your resources into protecting from black swans, you will go broke. If you prepare for only the rosiest circumstances, you will go broke.

These forecasts are not intended to be hard targets, that is why they are presented as a probability distribution. But they are intended to accurately portray what the realistic probability is for a certain outcome.

I missed by less than 1%.

chodpaba, I think your formulas are skewed because of the major disruption in June. I will stay with my prediction, next difficulty will be 1,8mill, to be more exact 1,798.900 I think network will have a very hard time hitting even your lower quartile at 1,82 let alone median at 1,98. There is almost no way network could grow 18% in the next 11 days. I understand that my prediction at 1,8mill again falls within your range, but your full range from 1,75 to 2,22 mill is just way too wide to have any meaning at this point. We are only looking up to 14 days ahead and your full range from lowest to highest is 30%.

If your next target has a range of 30% there is no way EVER you could predict next re-targets with any accuracy at all, let alone 4 targets out.

[skyhigh]

This is positive and negative criticism.  All this statistical terminology and formulas are nice but not very effective. Not so far back you had negative number for 139104 block. Around 1.56 mill, now you have it up to 1.93 mill. You keep tuning your formulas but the results are just way off target.

Natural growth of the network is between 6% to 12%. Unless something major changes in bitcoin usage which would of course dramatically change its price, network will grow at its natural pace of 6 to 12%. For major disruption for this to change price needs to raise or drop more than 50%. If it would drop 50% network hashrate might fall within -5% and +6% in case of price raising 50% network could grow from 6% to 25% depending on what the actual major change would be about. It's that simple.

Everything else is just some mumbo jumbo useless statistical formulas that so far look as accurate as throwing a dart. Your NEW 4 targets will all miss by a lot  but will still fall well within your widest range possible. Anyone can predict that. My prediction of 1.7  1.8 and 2.0 will be closer and it took me less than a minute to figure it out.

[gls90a]

I don't stop by and post much..... I'm one of those long time readers that does not get around to posting much.

Anyway, just wanted to say that I like looking over your extended difficulty forecasts. Its definetly not a perfect system but at least your taking the time to try and refine your approach and even willing to admit your mistakes.

5.4% variance on a market as volatile as bitcoin is pretty good in my book.

[skyhigh]

I was talking about natural growth of the network, at times where there is no major price change. You did the math for the most volatile month and a half bitcoin ever witnessed. This is why 6 to 12% in that case wouldn't work at all. Still way better than your crazy 3 +mill  prediction in post #48 and later on network drop.

Anyways, I'm not here to argue or anything. My initial post in this thread was that I disagree on predictions, and I stated my own 3 predictions, of 1.7mill 1.8 and 2.0 mill.

[2112]

Given that, your forecast for block 139104 would look like this:
Lower Extrema:   1792360
Lower Quartile:   1817724
Median:   1843088
Upper Quartile:   1868451
Upper Extrema:   1893815

I don't have a beef in this discussion, but I really wanted to comment on the way you present the results of your statistics. You are doing this in a really dilettantish way. What skyhigh was proposing was something like:

1.84mil (give or take 0.03)

Skyhigh is actually even more clever, since he talks about only two significant digits.

Your constant talk about quartiles and other advanced stuff combined with your unwillingness to share a model puts you in one of two classes: (1) you either deliberately lying or (2) you completely misunderstand what the statistics is all about and doing an arithmetical equivalent of making predictions from the tea leaves left in a cup.

Statistical theory is one thing, presenting it properly is the other. And I wrote here because I want to care about your results.

[nazgulnarsil]

my prediction is more accurate than yours.

It's going up. 

Behold the power of the austrian school of economics with its unflappable qualitative predictions! 

[hawks5999]

bro, you are feeding the trolls on your own thread. you'd be better to ignore them I believe...

[2112]

Chodpaba, I'll make one more post. From the picture above I can't figure out whether you are controlling the gnumeric or the gnumeric is controlling you. I see a lot of "tea leaves" on that picture, but some significant digits too. I noticed that you are most likely using Linux, so I have the following one-liner for you to run:

Code:

echo "scale=67; for(i=1;i<68;++i) 1/2^i" | bc -lq

I hope you'll recognize which digits are significant and which are "tea leaves".

I don't know why you wrote this thread and what is the purpose of your computation. All I can suggest is to get some better tools: (1) http://www.r-project.org/ (2) go to CRAN (3) download the main release (4) also from CRAN get the package "Zelig: Everyone's Statistical Software" (5) from the Zelig's website download its introduction: http://gking.harvard.edu/files/z.pdf .

Zelig will be as objective as it takes, and the knowledge you'll gain will be yours for the rest of your life.

[nazgulnarsil]

lol I never thought of it that way but I guess the austrians are the rolls of the economics branch of science.

[skyhigh]

Why you bother with 4 re-targets, if you always change them at least once before they happen. My dart throwing 1.7 1.8 2.0 mill will be closer on average than any 3 you will ever be able to calculate and NOT CHANGE at least once. So far you predicted / calculated re-targets but you soon discovered a better model or something else came up and you came out with new and new numbers. Calculate 3 re-targets and stick to them. I predicted my 1.7 1.8 and 2.0 on July 9th and I'm sticking with it. You changed your July 9th prediction at least 6 times by now. If I didn't write this or if you won't change the way you do your thing, you will keep adjusting and changing your future predictions.

[hawks5999]

Why you bother with 4 re-targets, if you always change them at least once before they happen. My dart throwing 1.7 1.8 2.0 mill will be closer on average than any 3 you will ever be able to calculate and NOT CHANGE at least once. So far you predicted / calculated re-targets but you soon discovered a better model or something else came up and you came out with new and new numbers. Calculate 3 re-targets and stick to them. I predicted my 1.7 1.8 and 2.0 on July 9th and I'm sticking with it. You changed your July 9th prediction at least 6 times by now. If I didn't write this or if you won't change the way you do your thing, you will keep adjusting and changing your future predictions.

Seriously, you've made your point a dozen times. You disagree with OP's method. Fine, move on.

[BTC_Junkie]

I apologize of this was asked before, but are you combining NMC and BTC total power for the forecasts? It looks like there was a spike when people switched back from NMC a few days ago when the difficulty increased.  Its possible that the combined rate is not significantly changed, just the focus of which blockchain people are on.

[geek-trader]

I know, it looks funny with all the up-and-down business. But the thing to realize is that the current Difficulty is affected by events in the price history as far back as three months prior. For this forecast the re-target at block 141120 is based only on price history from May 10, 2011 to August 1, 2011. The next three re-targets in this forecast include data from Time Series Analysis price projections: https://bitcointalk.org/index.php?topic=22685

The thing I am looking to extract from the price projections is the average daily low price for the relevant Difficulty retarget. I have found the average of daily lows to be the most useful segment from pricing distribution for doing the Difficulty forecasts. The prices shown below the relevant forecasts are the average daily low price projected for the Difficulty period immediately prior to the re-target.

In any case, the accuracy of the price projection will have a big effect on how useful the Extended Difficulty forecast is. I have not actually done enough testing to accurately represent the expected error for the 2nd, 3rd, and 4th re-target forecasts, this is something I am still working on.

Block Number:   141120
   
Difficulty   
Lower Extrema:   2077362
Lower Quartile:   2282446
Median:   2388374
Upper Quartile:   2515568
Upper Extrema:   2710510
2011-Aug-11



Block Number:   143136
   
Difficulty   
Lower Extrema:   1987536
Lower Quartile:   2129575
Median:   2208936
Upper Quartile:   2325841
Upper Extrema:   2462632
2011-Aug-27
$12.855



Block Number:   145152
   
Difficulty   
Lower Extrema:   2321853
Lower Quartile:   2455972
Median:   2507771
Upper Quartile:   2582639
Upper Extrema:   2697588
2011-Sep-08
$12.746



Block Number:   147168
   
Difficulty   
Lower Extrema:   2135109
Lower Quartile:   2257329
Median:   2304606
Upper Quartile:   2416883
Upper Extrema:   2598353
2011-Sep-22
$12.554

There has been some controversy about your predictions, but I, for one, apreciate them.

[hawks5999]

There has been some controversy about your predictions, but I, for one, apreciate them.

Thx.

Predictions are often controversial.

So you say, but we see how accurate you are

j/k

[Grinder]

But the thing to realize is that the current Difficulty is affected by events in the price history as far back as three months prior.

Or maybe it just looks that way at the moment because of coincidences. If you check for enough patterns you are bound to find some that fit, but that doesn't mean they'll continue to fit in the future.