For historical record, here is the original post of this thread.
tl;dr.* This is an analysis of previous crashes and what they might tell us about the next crash.
* I looked at crashes 45% or more, because I don't think it makes sense to attempt to sell in a smaller crash.
* Only 4 such crashes so far. Here are my
guesses (really guesstimates) based on these 4 crashes.
* The trough after the next crash should be at around 3,300, with anything from 2,300 to 4,600 being plausible.
* The next peak will be
at the very least 4,200.
Don't sell before 4,200.
* Once we reach 3,300 on the way up (the likely trough after the crash), the peak will happen anywhere from 14 to 27 days from then on.
Don't sell for at least two weeks after we get to 3,300. * Here are predictions for the dates. Read more below for details and limitations.
* Peak: mid-January.
* Trough: mid-May.
Background. This is an analysis of previous crashes and what they might tell us about the next crash.
In my
original draft about crashes, I considered all price declines of 15% or more, even if they were not off the All Time High. I restricted my
second analysis to only ATH's. I then saw an
interesting post by JulieFig and decided to drop the ATH restriction again.
A quote from JulieFig.
I simply defined a peak as a new ATH. However, this is assuming the magnitude of the peak that occurred in June 2011 is an outlier in terms of datapoints, but the date at which it occurred aligns with the proceeding peaks. Excluding this June 2011 peak magnitude, the peaks following (Jan and August 2012) could be considered ATHs. It is an assumption, yes, but a valid one (imho) considering the smaller user-base in the early years, and arguably less reliable exchange data.
The crashes so far. These are crashes of 15% or more.
pk.date pk.dow pk.price tr.date tr.price drop ft.date ft.price n.dest n.be
1 2010-07-19 Mon 0.09 2010-07-24 0.05 44.4 2010-07-27 0.06 NA 83
2 2010-11-07 Sun 0.36 2010-12-10 0.19 47.2 2010-12-11 0.22 11 68
3 2011-01-16 Sun 0.39 2011-01-19 0.31 20.5 2011-01-20 0.35 10 5
4 2011-02-14 Mon 1.06 2011-04-05 0.67 36.8 2011-04-07 0.75 14 62
5 2011-05-14 Sat 7.86 2011-05-21 5.97 24 2011-05-23 7.1 2 11
6 2011-06-09 Thu 29.6 2011-11-18 2.14 92.8 2011-11-24 2.42 42 617
A 2012-01-08 Sun 7.05 2012-02-16 4.19 40.6 2012-03-26 4.65 11 184
7 2012-08-17 Fri 13.3 2012-08-19 9.09 31.4 2012-08-23 10 18 110
8 2013-04-09 Tue 215 2013-04-16 65.3 69.6 2013-07-08 77 17 209
B 2013-09-03 Tue 145 2013-10-02 110 24.5 2013-10-04 121 21 45
9 2013-11-30 Sat 1,130 2014-04-11 392 65.4 2014-05-07 441 17 NA
C 2014-06-03 Tue 668 2014-06-14 560 16.2 2014-06-30 625 10 NA
* Crashes from an ATH are labeled with a number. Those not from an ATH are labeled with a letter.
* 12 crashes in all. 9 crashes from an ATH.
* pk = peak, tr = trough
* ft = "foot". According to my simplistic algorithm, this is when the next move up begins.
** Trough and foot for the current crash(es) are provisional.
* n.dest = number of days of gains (prior to the peak) that were destroyed in the crash.
* n.be = number of days to break even if you bought at the peak.
* Take a look at the days of the week for the peaks. It's mostly the weekend or thereabouts.
What does this mean?Fri Mon Sat Sun Thu Tue
1 2 2 3 1 3
Are 15% crashes all that interesting? Can they even be considered crashes? Given how volatile bitcoin is, I was actually surprised that it only fell by 15% or more a mere 12 times.
In a previous discussion, I figured out that, for many people,
the expected drop would need to be 65-70% for them to sell. For a risk-tolerant person who is paying taxes in the US, the expected drop to sell has to be at least 55%.
Let's look at the crashes of 45% or more. Because if a crash is less than that, it's probably not worth your while to sell.
Do you agree with this? Is this good reasoning? pk.date pk.dow pk.price tr.date tr.price drop ft.date ft.price n.dest n.be
1 2010-11-07 Sun 0.36 2010-12-10 0.19 47.2 2010-12-11 0.22 11 68
2 2011-06-09 Thu 29.6 2011-11-18 2.14 92.8 2011-11-24 2.42 42 617
3 2013-04-09 Tue 215 2013-04-16 65.3 69.6 2013-07-08 77 17 209
4 2013-11-30 Sat 1,130 2014-04-11 392 65.4 2014-05-07 441 17 NA
* Only 4 "real" crashes so far.
Predictive ratios and durations. Let's see if we can find some ratios and durations that repeat consistently enough crash after crash.
pk.date pk.price r.tr.pk1 n.dest n.tr.tr n.ft.ft n.pk.tr
1 2010-11-07 0.36 NA 11 NA NA NA
2 2011-06-09 29.6 5.94 42 337 342 181
3 2013-04-09 215 2.21 17 515 592 508
4 2013-11-30 1,130 1.83 17 360 303 228
* r.xx.yy = ratio of xx to yy. n.xx.yy = number of days from yy to xx.
* r.tr.pk1 = the trough price of the current crash divided by the peak price of the
previous crash.
* I could not find a price ratio for the peak that was consistent enough.
Ideas?Based on these 3 or 4 observations, here are my guesstimates for the future values.
name p_25 p_50 p_75
1 n.tr.tr 350 397 450
2 n.ft.ft 324 394 480
3 n.pk.tr 205 276 372
4 n.dest 13.8 19.1 26.6
5 r.tr.pk1 2.03 2.88 4.09
* Let's start with a prediction for the trough after the next crash, using r.tr.pk1. The trough should be at around 3,300, with anything from 2,300 to 4,600 being plausible.
p_25 p_50 p_75
1 2,300 3,270 4,630
* I could not find a ratio for the peak that was consistent enough across these 4 crashes. However, since we are looking at crashes of at least 45%, and since we have guessed the troughs, we can calculate the
minimum peaks based on that.
The likely peak will be at least (that is, if the crash is only 45%) 5,900. The lowest plausible minimum peak (that's a mouthful) is 4,200.
No matter what, don't sell before 4,200.
p_25 p_50 p_75
1 4,190 5,940 8,420
* For n.dest, the prediction is given above. The next "big" crash (45%+) will destroy about 19 days worth of gains. Anywhere from 14 days to 27 days is plausible.
Since we've estimated the trough, the other way of looking at it is like this. Once we reach 3,300 on the way up (the likely trough after the crash), the peak will happen anywhere from 14 to 27 days from then on.
Don't sell for at least two weeks after we get to 3,300. * To make predictions based on the other durations, I am using a truncated distribution. This is because we know that these events will occur after today -- this gives us extra information. This also means that the predictions below will change as more time passes.
name p_25 p_50 p_75
1 Peak (based on n.pk.tr) 2014-11-03 2015-01-12 2015-04-17
2 Trough (based on n.tr.tr) 2015-03-26 2015-05-12 2015-07-04
3 Foot (based on n.ft.ft) 2015-03-27 2015-06-05 2015-08-30
* The peak is likely to occur sometime in mid-January. It's plausible any time between early November and mid-April.
* The trough is likely to occur sometime in mid-May. That's 4 months in a bear market. Could be any time between late March and early July. So could be half a year or more of bear market.
* The foot -- the beginnings of the next move up -- could be some time in early June. The foot has to come after the trough, and
the above prediction does not account for that. This is a weakness of this prediction.
Previous version. 
