See update ⊜ at bottom after reading the description.
When binned over 1-week intervals, we easily see the RSI (Relative Strength Index) fall dramatically, indeed nearly vertically, at the burst of each 'bubble', as expected when viewed in this way. I suggest that one should make a note of the first significant recovery segment after these drops —as indicated by a relatively sharp/discontinuous change at one end, creating a region of convex interior— and record the relative percentage change over which it occurs: call that "A".
The next time a relative percentage change is observed ("B", say) to be greater than or approximately equal to the inverse of "A", (i.e. B ≳ -A), where it is again bounded on at least one side by a semi-discontinuous change —up, down or sideways; it does not matter— forming another mostly convex region, (as in the case of "A")— I claim that this will signal the bottom of that short-term crash and thus begin the interlude before the next 'bubble' inflates; and that, furthermore, the RSI (1W) will not again fall beneath the threshold set by the most recent recovery region, (e.g. ~42-48, in this most recent case.), for at least 23 weeks hence, or until after a new bubble bursts, whichever comes first.
Because my hypothesis also involves the convexity of the regions as a premise, it is possible this most recent recovery is not yet complete, as it does not strictly adhere to that assumption. However, my overall point is flexible enough to not be needlessly entrenched in that aspect. Also, the aforementioned features are all relative so, while I use RSI (14, close), other parameters revealing similar structure may be used as well, though I've not yet explored the merit of such variations.
⊜ Update: For a more expansive view of this concept, including implicative evidence regarding the recent bubble period offset correction, see my newly published chart I've linked to this one "INTRA-BUBBLE LOWS PERIOD ... " — an early preview of which can be see in the chart image here or an even earlier version (same basic content) in the comments section. — If you were linked here from the other than you already know all of that anyway.
Note: As with all my charts, these findings are statistical. I do not pretend to have any insider information or special knowledge of market psychology which should contribute additional meaning to these observations.
I've given an expanded treatment of the phenomenon in this chart image, including periodicity, which also appears to hold if ones looks at Huobi (and other) data [not included]:
The period between pre/post–bubble bottoms appears to be strongly consistent at about [36 weeks]. The most recent small deviation from that also being intuitively explainable as resulting from the false-start recovery blip, as indicated in the chart image as well.
snowsilence
⋅
For anyone who has not yet seen that this idea has been hugely successful, see the updated chart image posted below, as well as the counterpart to this chart and the implications therein, which is located at — tradingview.com/v/VKcP7LDt/ — and also includes a link to a 'live' version of the chart.
Morbius
⋅
Interesting... A few comments/questions, if you allow:
1) I'm not entirely sure why you need to define the condition as "if B>=-A". It seems A can also just be defined as "the first recovery after the near vertical drop", in which case you're saying "reversal if B>A", which is basically saying "we reverse if we ever see a RSI upwards move larger than the one immediately following the first crash after reaching ATH". But correct me if I misunderstood somethign.
2) Did you test if it holds for the extended trading history via mtgox? Especially the 2011 bubble cycle comes to mind. The old history can be (manually) added by setting the price data to come from: "mtgox:btcusd"
3) I like the method, and it makes sense, but like you probably realize yourself: 2 (maybe 3) data points don't lend much statistical power to the method. That's of course a problem with *all* long-term indicator signals in BTC price history (like weekly EMA crossovers)... we simply don't trade very long yet.
In any case, I appreciate the observation and the post.
snowsilence
⋅
1.) I say B>=-A because I've chosen to call the label the first regions as (-)negative percentage changes —thus A is negative already— and the B's as (+)positive ones, as you can see from the chart, which is merely to aid in visual intuition. The condition is equivalent to |B|>=|A|, where the bars indicate absolute values.
2.) I have tested it with the historical Mt. Gox data, where it also holds; as well as with Huobi, in case anyone's interested.
3.) As stated in the note at the end of the above description, I'm not even claiming that this always works or makes any sense.
The period between pre/post–bubble bottoms appears to be strongly consistent at about [36 weeks]. The most recent small deviation from that also being intuitively explainable as resulting from the false-start recovery blip, as indicated in the chart image as well.