The unknown

Assumptions of the unknown

Why the unknown? Well, lets make a BIG difference between "The unknown" and "The unknowable". This isnt purely my distinction, its from shaman and occult traditions. Trading is either dealing with the unknown OR the unknowable. If its a matter of the unknowable, well then if thats correct its a futile project to be trading, since you wont ever know what to do, since its UNknowable, it cannot ever be known. But then if trading is a matter of the UNKNOWN then it MIGHT become known, only that at the moment its not.

So we need to assume we are dealing with the unknown. The next assumption to make is about certainty. Are there any in trading? To me the only certainty is that it will end rather soon. The whole thing. Industrial society. Combustion engines. Oil. Money. All of that is quite tenuous, quite unstable, quite insane and speeding towards its end. If you would like to discuss this, we can do that, but lets instead move ahead and get technical/mathematical.

The assumption that there are no certainties in trading is also a matter of theoretical probabilities. Well all probabilities are theoretical, but anyway. This is so in such a manner that we might WANT to try to deduce or induce or in other manners try to know a certain probability. I dont like purely theoretical examples so im grabbing one from reailty. Here is my small position right now

Attached Image (click to enlarge)

:

Like so. This assumption means i dont know how big the probability actually IS that my position will hit TP and how bit it is that it will hit SL within say 2 days. I really DONT know. It might even be a unknowable fact what the probability is.

What i sort of do is look at this:

Attached Image (click to enlarge)

Remember, because there is no probabilities to be KNOWN, i cannot say that i will probably get my TP. Meaning the actual probability that i will win, is unknown, maybe even UNKNOWABLE.

Because of this assumption, there IS no systems such as any combinations of indicators or any rule such as "buy the lows" and "sell the highs". This is not purely theoretical, i have done the calculations and tried it all, everything anybody could think of including me.

But how then is one to operate, in short, how does one trade? I think the trick is to trade WITH the unknown. Trading with the assumption that we are dealing with the unknown. That DOESNT mean that the unknown is UNSYSTEMATIC, only that the unknown isnt KNOWN. Further more, the unknown doesnt mean its inoperable or UNSYSTEMATISABLE or UNSYSTEMATIZED.

Let me come at it from a slightly different angle.

Trading is a matter of the unknown. The unknown is not known at the moment. Irregardless of how much price is studied, no probabilities will ever be known about it. This is why it sort of often seems "random" to a lot of people. Yet its not random. It IS systematic. But we cannot EXPLAIN or actually KNOW or actually PUT IT INTO WORDS in what way it is systematic.

In mathematical, computational terms it takes a super human mind to calculate where to enter, where to put TPs and SLs. In this manner it is UNKNOWN how to trade. I think that if you would ACTUALLY be able to put trading into words or other symbols like mathematics, then you wouldnt want or need to do it anymore. You might as well just do something ultra weird like calculate where someone is about to accidentally drop some cash, stand there and just pick it up when they drop and just walk away with it.

But we can approach the unknown and the push the boundaries of what is unknowable. What once was unknowable might become unknown and the unknown can become known. Thats because there really are no probabilities. Otherwise we might be able to calculate what was unknowable and then we might as well not try to do that since it was UNKNOWABLE.

Yeah well anyway, i am quite sure i wrote the wrong thing a couple of times. I might also want to scrap this whole thing and cross it out as a mistake. More is to come i feel.

Some further assumptions
Well other then that there are a few more assumptions i have made after a lot of trial and error and experience and so on.

Trading needs to be sort of fun.
If its altogether boring, it wont work. At least not for me. At several occasions i have been really convinced i knew how to do it. And then i loose like 5% because i was so sure of how to do it. It needs to be sort of fun and i need to keep guessing A LOT. As soon as i think i KNOW whats going to happen, the market Whops my ass something FIERCE. I mean since im HELL bent on the fact that i KNOW, i can add to a position or just WAIT for my TP to happen. Guess how that ends eventually? Well maybe if you dont have the answer you need to experience it for yourself a couple of times.

I am really interested in the unknown.
I was always interested in the unknown, ever since i was a kid i was always into more or less weird stuff. But i never really got stuck on anything. Also i have A LOT of interests if i need to count them. But maybe its all about the one interest in the unknown. For instance i have played music with a band. It was fun as long as i didnt know what what going to happen. When i figured out the whole thing, it stopped being fun. I still miss it but if i do it again it needs to be in another setting where i need to keep guessing whats going to happen again. Same thing with trading. Although trading is actually better that way since i probably ALWAYS have to keep guessing.

If the price is the output of a possibly deterministic yet unknown generating process then, per definition, it is a random variable.

I can show you that it is wrong if you like. (but I don't want to spend 2 hours to do it if you're not interested)

I am interested in this.

Well, i think that it is unknowable if its deterministic or not. At the moment. For me to say its deterministic i would have to SORT of know the equation that makes price, right? And almost the same thing goes if if was to say its NOT deterministic because i would have to know that price is NOT an equation, meaning i would have had to collect all the variables and make the computation and see that it does not equate to price. Correct me if im wrong...

About the second thing, yeah i would like to see how you show me that there are probabilities to price, that matter in actual trading. Even though i am sort of convinced that you are correct already i still would like to know how you think.

As are the whole question of determinism and ACTUAL probability. That doesnt mean non determinism or even some kind of personal beef religious thing either. I mean a "all knowing" state of mind is a prerequisite for having a experience based view on determinism or non determinism.

[Sorry long post ahead]

I once made an experiment with an oscillator to find a way to buy the dips (spoiler: I failed!). I'll try to reproduce the thinking process. For this post I'll use the RSI because it is a well known standard indicator. The period is the default 14. I'll use EUR/USD because everybody trades it. I wanted to use the daily chart but the datatset is too small. I'll use H4; with lower TF the fat-tail effect becomes too important to see what I'd like to show.

Place a RSI14 on your E/U H4 chart. You'll notice that when the market trends up the oscillator is pushed to the upper area. When the market trends down the RSI is pushed to the lower area. This implies that this oscillator is never oversold in a uptrend. I needed correcting this. A simple way to do it is to substract a SMA of the oscillator from it. The period needs be at least twice the original period to catch the main cycles. 2x14=28 I rounded to 30. RSI oscillates around a non-null mean (level 50). Of course substracting a MA will remove this constant so I add 50 to the result to restore the range. The formula of the detrended RSI is "RSI - SMA(RSI) + 50". This is the result (red=RSI, blue=detrended RSI). The blue one is more centered. It's not perfect of course.


An oscillator will cross its center line either from above or from below. It can make a top or a bottom. This can happen above or below the center line. A top/bottom is defined as a sequence of three values the second being greater/less than the two others. Let's take these 6 "events" and let's split the data at each occurence of such an event. Between two events the price moves. This movement is negative if the price goes down. I associate the set of all the moves to their respective event. The idea is to get a probability distribution of the future move given an event. Here are the 6 histograms. In red I fit a gaussian distribution. The vertical line is the mean.


We can see different things:
- The distributions look like as if they were made of the sum of a gaussian and another much more skewed distribution.
- The mean is sometime positive sometime negative but always very close to zero.
- The variance is huge.
Is the small mean significant? I repeated this with AUD/USD and GBP/JPY. For each event it is either positive or negative. So no.

Such a looong post to say there is nothing?!
Remember that we detrended the RSI. On the picture you can see that the re-centering makes the original RSI above the corrected one when the market trends up. Conversely for a down trend. Let's rebuilt the histograms but this time only selecting the cases where the RSI is above the corrected oscillator and one more time when the RSI is below it. We now have 12 events/histograms. I'm not going to post the picture because it looks really like the previous one. Instead here is an excel screenshot of the value of the means for the 3 instruments (values are in pips --real pips not Metratrader's!--):


You may notice that when the trend is up the values are almost all positive and when the trend is down the values are almost all negative. For the three instruments. Without considering the trend it is alsmot half-half (7/18). This cannot be by chance. Getting only 1 positive value out of 18 random numbers with mean zero is like one head out of 18 coin flips. The probability is 0.0069%.
Note: in my original experiment I used a smoother oscillator. The means were more pronounced and the distributions with a zero mean were more skewed in the direction of the trend.

If we only considered one of these histograms any usual test statistic (https://en.wikipedia.org/wiki/Test_statistic) would have been unable to reject the null hypothesis that the mean is zero! There is something but it is burried in noise. In case you're skeptical this is the values you get when you select the cases where the difference between RSI and the detrended version is >20:
downCross -16.8
upCross 5.07
topAbove 29.1
topBelow 11.1
bottomAbove 15.2
bottomBelow 0.508

How do you explain the negative value for the event downCross? What happens is a fall of the RSI = price drops below the average of its range in a strong up trend => this is a reversal! Remember I use a SMA30. It lags 15 bars.

Conclusion:
By studying the price history we found that:
- RSI generates very poor signal. I can explain why but it is out-of-scope in this thread.
- Trends clearly exist and provides some edge even with a 15 bar lagging indicator.
- The noise is so huge that the expectancy produced by the trend is inexploitable short term.
- Because of this terrible signal-to-noise ratio test statistics are inefficient with market data. This explains why the pure random walk assumption is easily accepted.

You might also be interested in these two posts of mine:
https://www.forexfactory.com/showthr...04#post8464204
https://www.forexfactory.com/showthr...10#post8466810

TL;DR data says when price is very overbought... buy!

Thanks for this Pip. I could handle your conclusions but you lost me with your detailed analysis.....that's why it is great to refer to the summary abstract....while you do all the hard work/tic

Do you have a thread where you elaborate on this statement? I didn't realize MT4/5 used its own definition of a 'pip'? I thought a pip was a fairly standard, concrete, 1/10,000th (or 1/100th of a yen) of a currency unit?

I've now read those two threads; very interesting but I think I got more from your pithy summary. Could I restate it thus (and still be accurate/precise)?

"A price trend making new highs will tend to continue making new highs, while a price trend making new lows will tend to continue making new lows (all other things being equal)?"

Also is this a rejection of the 'regression to the mean' market theory?

Yes as far as I could test it FX is not following a Ornstein–Uhlenbeck process even during a ranging period. Of course it oscillates around some level but nothing "pushes" it back to the mean. It seems to follow an integrated Ornstein–Uhlenbeck process. I mean the trend is mean reverting with a theta around 0.99

Good one!!

You're right of course. I think he refers to metatrader's profit/loss column which show tenths of pips. So 123 in p/l is 12.3 pips.

I would rather see a 12 turn to 13 any day. But there are a lot of folks who would say, "I just made 123 pips, Yaaaaaaaaaayyyyyy"

Good times man

Ah yes, points, or 'pipettes'. A scam sold as improved precision.

I've done some reading about O-U process, so naturally I'm an expert.

Are you saying that with theta ~ 0.99 the total non-reverting drift is only about 1 percent? I would be happy to trade reversals from the top and bottom ranges of my chosen trading horizon with only 1 percent failure to worry about.

Hi Pip:

In MHO the Tracy-Widom Distribution would help to explain the results form your analisys, see the next article:

https://www.wired.com/2014/10/tracy-...atistical-law/

Will be appreciate very much yours comments
Thank you in advance.

Sorry. I meant that if the trend is X at this bar, the trend at the next bar will be 0.99X plus noise. The reversion is weak.

I didn't know about this distribution. What make you think about it?

I actually do resonate with this a whole lot. There have been many times where I was so convinced that I just figured it all out - only then I was proven I hadn't got it all after all. Lately my findings derived from my own observational and experimental data suggest that the times where we feel like we're invincible that we truly know with certainty we've got this is not entirely an illusion that there's some kind of truth. You have found a pattern, combinations of conditions you try them out, works great over few trades all winners. Then you go ahead and try them out on your live account or demo some more and it seems to stop working. What I've found out is that it actually doesn't stop working. You start doing something else, for instance your mind (maybe its your ego) starts associating all kinds of variations of patterns the same as your initial profitable pattern. Our mind is messed up. I believe markets are not random. There are times where it can be random. But we are good at spotting things that aren't random to begin with. Thats our biggest advantage. But when its about to take off our mind (Ego) gets in the way I think. A great way to combat this is I think to have focus if you trade multiple pairs maybe you only focus on 1 or 2 patterns forget about the variations they'll only confuse you and drain your account balance , or focus on one pair but you trade them in all conditions.

Dear PipMeUp,

I tend to agree with ADance on this and I believe this quote from the referred article would explain why:

Quote

Disliked

Whereas “uncorrelated” random variables such as test scores splay out into the bell-shaped Gaussian distribution, interacting species, financial stocks and other “correlated” variables give rise to a more complicated statistical curve. Steeper on the left than the right, the curve has a shape that depends on N, the number of variables.

Basically we all agree that trends do exist and during a trend the distribution of downward & upward price moves do vary. For example, in a strong uptrend the maximum upward move can be 4-5 standard deviations away from the mean, whereas downward moves tend to die out around 1-2std away from the mean. For an ordinary guy like me, this makes it too difficult to pinpoint "downward out liners". If I could, then I would have applied a scaling up logic - eg. if uptrend then buy when the price falls below x stdev below the mean - but I can't do it with Gaussian rule book. Standard deviations fail me...

Price changes are correlated and having learned that Tracy-Widom Distribution is a better fit than Gaussian, I seek for a better answer: I really wonder how one can use this in the practical trading strategy above? What would be the replacement of "x stdev"?

I would appreciate hearing your thoughts on the matter.

Tracy-Widom distribution is defined over Hermitian matrices. In the context of trading I can only think about a correlation matrix (which is Hermitian). It may then be useful for portfolio momentum investing but I can't figure out a use of it in the context of trading a single asset.

BTW the second order moment (variance) is defined for this distribution.