Experimental indicator, anyone? I give you the... reversal index

[oda.krell]

(EDIT)

Please see my posts further down this thread for the recent values of the indicator and my best suggestion how to interpret the output. Consider what I wrote below the following horizontal line obsolete.



Been playing around with ideas to (mechanically) quantify the likelihood of short term trend reversals (or trend continuations). The method I came up with is defined well enough that I could in principle fully code and backtest it, but tbh, for now it's just easier to do the last step (which requires identification of some "dominant" trend in the recent past) manually. So, no backtesting yet, but instead: seemingly random numerical values! Everyone loves those, no?

I might go into more detail how I calculate the values later, but for now I'll just say that the calculation is based on some short-term comparative analysis of price trends. The value you get should be thought of like a classical oscillator:
values near 100 = certain continuation(edit: wrong way to describe the top value. The method is more reactive than predictive, so the correct way to talk about a high value is) no sign of reversal.
values near 0 = certain reversal,
values near 50 = outlook hazy, ask again tomorrow.

You might wonder what the "reversal" time frame is I try to predict here... I hope to pick up a reversal or continuation of the currently dominant short term trend I determine in the first step, which happens to be in the range of some weeks. In EW terminology that would be a 'minor trend' I suppose.

I'll try to post a daily update, unless I lose interest... Alright, first value:


2014-06-04, 3:08 PM CEST

CI = 54
current main trend: up
price at calculation: $636


(edit) to have some basis to track performance, I'll add the approximate price at the time of the calculation above... median price of the last closed 2h interval, and the direction of the main trend that I track, i.e. the trend of which I try to determine whether it will continue or reverse.

(edit 2) clarification of the above: the main trend I track in calculating the CI value is  currently 'up'. The CI value expresses the likelihood that the dominant trend continues to be up. At 54 right now it basically says: "Unclear situation, but slightly leaning towards (upwards) continuation".

...


Yeah, I know. Not exactly enlightening, but that's what I got.

[Bitbuy]

Hi Oda;

Definitely looks interesting, and since it's coming from you, I'll keep an eye on it
Will your updates always be at the same time?

[oda.krell]

Hi Oda;

Definitely looks interesting, and since it's coming from you, I'll keep an eye on it
Will your updates always be at the same time?

Don't think I'll be able to do that. And I don't think it really matters... what ideally would happen is that the CI could help figuring out where we're heading if we are in one of those "in between" states that last for some days, i.e. not the shorter periods (think: flash crashes), nor the longer periods (think: bear market or bull market?). And for the time frame I have in mind, 1 or 2 hours early or late won't really matter much, I think.

[Peter R]

Interesting.

So if it works, you'll have us convinced because you've posted the predictions ahead of time.  However, we still won't know how to calculate your indicator.  Which means you can charge us top dollar for a subscription to your newsletter

[Bitcoin_is_here_to_stay]

Oda, so currently "continuation" means more creeping up? That should be also clarified at the moment you post the values.

[oda.krell]

Interesting.

So if it works, you'll have us convinced because you've posted the predictions ahead of time.  However, we still won't know how to calculate your indicator.  Which means you can charge us top dollar for a subscription to your newsletter

Ha, I wish.

If the thing is even remotely informative, I'll happily share the algorithm. I'd probably need some help anyway to fully code it... the tricky step I mentioned, 'determination of dominant short term trend' can probably be automatized through some combined analysis of BBW and ROC, but it's probably not quite as simple as looking at hard coded threshold values.

So when I say "I could code it in principle" I really mean "a more competent programmer than me could code it probably"

[oda.krell]

Oda, so currently "continuation" means more creeping up?

Yes. The CI value has to be seen in relation to the direction of the dominant trend. That trend is 'up' and CI value is slightly above 50, so the result of today says: unclear where we're going, with a slight bias towards continuation upwards.

EDIT: you're right, that wasn't very clear. I re-organized the presentation. Better now?

[sgbett]

I might not do TA very well, but I can code up a "hello world"

[Bitcoin_is_here_to_stay]

Oda, so currently "continuation" means more creeping up?

Yes. The CI value has to be seen in relation to the direction of the dominant trend. That trend is 'up' and CI value is slightly above 50, so the result of today says: unclear where we're going, with a slight bias towards continuation upwards.

EDIT: you're right, that wasn't very clear. I re-organized the presentation. Better now?

Yes, thanks! Especially if we would like to revisit a couple weeks from now it is very helpful to know what "current trend" estimation was at the time

[oda.krell]

Oda, so currently "continuation" means more creeping up?

Yes. The CI value has to be seen in relation to the direction of the dominant trend. That trend is 'up' and CI value is slightly above 50, so the result of today says: unclear where we're going, with a slight bias towards continuation upwards.

EDIT: you're right, that wasn't very clear. I re-organized the presentation. Better now?

Yes, thanks! Especially if we would like to revisit a couple weeks from now it is very helpful to know what "current trend" estimation was at the time

Are you suggesting that in a couple of weeks from now the dominant trend could be anything other than 'up'?!?

...

BURN THE HERETIC!

[oda.krell]

2014-06-04, 11:11 PM CEST

CI = 58
current main trend: up
price at calculation: $645

[Bitcoin_is_here_to_stay]

Oda, so currently "continuation" means more creeping up?

Yes. The CI value has to be seen in relation to the direction of the dominant trend. That trend is 'up' and CI value is slightly above 50, so the result of today says: unclear where we're going, with a slight bias towards continuation upwards.

EDIT: you're right, that wasn't very clear. I re-organized the presentation. Better now?

Yes, thanks! Especially if we would like to revisit a couple weeks from now it is very helpful to know what "current trend" estimation was at the time

Are you suggesting that in a couple of weeks from now the dominant trend could be anything other than 'up'?!?

...

BURN THE HERETIC!

LOL. I first meant "couple of months' but then was not sure how long you will keep it up. Good catch

[YipYip]

Interesting.

So if it works, you'll have us convinced because you've posted the predictions ahead of time.  However, we still won't know how to calculate your indicator.  Which means you can charge us top dollar for a subscription to your newsletter

Ha, I wish.

If the thing is even remotely informative, I'll happily share the algorithm. I'd probably need some help anyway to fully code it... the tricky step I mentioned, 'determination of dominant short term trend' can probably be automatized through some combined analysis of BBW and ROC, but it's probably not quite as simple as looking at hard coded threshold values.

So when I say "I could code it in principle" I really mean "a more competent programmer than me could code it probably"

I can code it for you if u want ?

[N12]

Does the indicator itself determine the main trend? Is it binary?

[oda.krell]

Does the indicator itself determine the main trend? Is it binary?

Not right now, but it's not a really controversial choice in what the main trend is imo. Like I said above, I think it could be coded by looking at, for example, rate of change. It's just easier to do by hand if you're lazy, like I am. And it is binary. Positive slope, or negative slope. Nothing in between.

I can code it for you if u want ?

I'll keep that in mind, thanks. First, I'd really like to see if this thing has any informative value though.

[oda.krell]

2014-06-05, 3:39 PM CEST

CI = 60
current main trend: up
price at calculation: $647

* * *

Please note the correction I made in OP: the way to look at CI values near the top is different from what I said at first... it does not amount to the claim that "reversal is unlikely", but rather "no indication that a reversal is underway".

So I should probably rename it "reversal index", and map the current 0 effect to 100.

[oda.krell]

Okay... as mentioned one post above, I changed my mind about how to interpret the output of my indicator, though I didn't change the formula itself.

I'd rather think of it as a 'reversal indicator', because the (previous) upper side of the value range doesn't really say that a continuation is guaranteed or even likely, just that no signs for a reversal can be picked up. Think of it like looking, for example, for bearish divergences on the RSI: finding one is an indication that a trend is about reverse, but not finding one is not equally an indication that the trend will continue. Below I'll re-post the previously calculated values as: RI = 100-CI, i.e. mapping the 0 effect to 100, and vice versa, and calling it "RI" (reversal index) from now.

Once more, the suggested (from now on) way to read the RI value is as follows:

100 = reversal (from main trend) certain
50 = undecided/unclear signal
0 = no sign of reversal (by the metric of the RI)

(edit) To keep track of performance, I take note of current price (H+L/2 of last closed daily interval at time of calculation), and the direction (up, down) and starting point of what is assumed to be the current main trend.

This is also the last new post I'll make in updating the values. all further values will be added inside this post by edits. I feel this method isn't really worth bumping to the 1st page of the forum every day. Feel free to check on it if you wish though, I'll keep updating it daily, if possible.


2014-06-04, 3:08 PM CEST

RI = 46
trend: up (from 2014-05-20)
price: $665


2014-06-05 3:39 PM CEST

RI = 40
trend: up (from 2014-05-20)
price: $648


2014-06-06 3:02 PM CEST

RI = 39
trend: up (from 2014-05-20)
price: $655


2014-06-07 3:07 PM CEST

RI = 44
trend: up (from 2014-05-20)
price: $655


2014-06-08 8:37 PM CEST

RI = 38
trend: up (from 2014-05-20)
price: $647


2014-06-09 3:00 PM CEST

RI = 45
trend: up (from 2014-05-20)
price: $656


2014-06-10 4:00 PM CEST

RI = 43
trend: up (from 2014-05-20)
price: $649

comment: RI continues to stay below 50, which implies the larger upwards trend is likely still intact. Momentum signals differ: 1d MACD just turned negative, so it'll be interesting to see how the situation resolves.


2014-06-11 04:15 PM CEST

RI = 49
trend: up (from 2014-05-20)
price: $650


2014-06-12 3:30 PM CEST

RI = 51
trend: up (from 2014-05-20)
price: $634

comment: first time during this swing that RI goes above 50. Slightly so, but if past results apply here as well, it's starting to look like the downturn isn't over yet.

also, I'm already not too happy with the results of the RI, but have some ideas how to fine tune it better. I will however continue this first "experiment cycle" with the old rules to get a result at the end that allows me to draw some conclusions. Afterwards, if I think improvements might work, I can do the next cycle with the new rules


2014-06-13 4:00 PM CEST

RI = 52
trend: up (from 2014-05-20)
price: $590

comment: I'm late with the post, but the value above is calculated for the chart ~24h after the last one for consistency.

The method doesn't really provide a price target, but if I look at the situations in the past where similar values came in, I would consider ~$530 to be a possibility in the coming days.


2014-06-15 2:00 AM CEST

RI = 57
trend: up (from 2014-05-20)
price: $566

comment: Last update for now. RI = 57 just states what is obvious anyway: the previously dominant trend is broken. I'll stop posting updates for now because I don't see how it can inform anyone's trading. We went down substantially, and the question right now is if, how and when we will go back into an upwards trend, but the RI in its current form can't answer that question.

I'll make another post with a more detailed analysis of the results for now, together with a chart of the values of the RI, and then I'll work on improving it. The RI value crossed 50 shortly before the crash began, but e.g. daily MACD would have actually performed better this time, so I should work on it.

[sgbett]

bump away! its better than some of the tripe that keeps being pushed to the top

[oda.krell]

Huh. That's a nice surprise...

I spent a bit of time backtesting the formula (still not all the way automatized, but I'm getting reasonably fast at it), and I took for example a look at the April 11 rally and the period that followed.

It's an example of a scenario that I was hoping my RI would be able to pick up: there's a strong trend, an initial retracement, followed by a longer period of indecision. In situations like that, I'd like to get a leading signal from my RI whether reversal is expected or not.

Here's the period I'm talking about:



Marked in green is the beginning of what I define to be the main trend, orange is the crash I'd like to predict.

Turns out, at the time of the last green candle *before* the crash, RI was 55, i.e. "undecided, leaning towards reversal" by my interpretation guidelines as specified so far. So the RI was leaning in the right direction (reversal signal more likely than no reversal signal), but the value was pretty close to 50. Perhaps the way to make use of it is considering even small deviations from that baseline as a relevant signal. Further testing needed obviously to determine if an "aggressive" interpretation like that gives too many false signals or not.

Anyway, the above "aggressive" interpretation applied to the current situation and values would mean that the values we're seeing now (high 30s, low 40s) are actually a comparably strong indication that there is no reversal signal coming in from the RI. (The updated values can be found 2 posts above)

Let's see.

[oda.krell]

Another example of my RI backtesting. This time on the early March (failed) breakout:



RI reversal signal came a solid 3 days before daily MACD turned red, and at around $40 higher price.


For comparison, 6h MACD turned red around the same time as RI signaled 'reversal' (slightly earlier in fact), but it gives a lot more signals altogether, including many bad ones, so can't really be compared:




EDIT: the current situation is actually quite interesting. Almost all momentum signals of comparable time scale are either red, or are about to (1d MACD will probably go down in the next day or two). My RI signal is still saying "doesn't look good, but no reversal case yet". Will be really interesting to see if it is just lagging even more than the momentum signals at the moment, or whether it is onto something.

I never properly defined what I'd consider a "success", but since I'm thinking in terms of profitable trading signals, I'd want to see at least a 5%, and preferably a 10% change in price from here to say that we substantially deviated from the main trend. In other words: staying above 618, or 585, in case RI doesn't signal reversal would be a success in my view. If price hits 585 before RI signals reversal, I'd consider it a clear failure.