The miracle of the grail... Myth or reality?! - page 12
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
This is where you get the concepts mixed up... :-)
The more degrees of freedom, the fewer the parameters...
More precisely, the fewer parameters, the more degrees of freedom the TC has
Well, no.
For example, let's take a system without explicit parameters - every day at the same hour we see what the previous bar was and open exactly one hour in the same direction. Such a system has very few degrees of freedom. There is no way to "tweak" it.
Now we start adding parameters to it - say, we open not for an hour, but for some time, set by a parameter. Plus, we will set TA and SL, which are also set by parameters. Plus - we will not look at one bar, but five bars, and enter depending on this five-bar pattern. In addition, we will set a filter with some indicator, and not only one. As a result - we'll get a system with a hundred parameters, which can be "adjusted" for any symbol, any timeframe, for a long period of time... But will it work? I'm more than confident that it won't - precisely because it has many degrees of freedom - each parameter will be fitted in the optimizer based on history, and it will work on history. In real trading it won't.
As I said before - for most TS, the only parameter that needs to be optimised is the size of the moving observation time window. Ideally, it should be self-tuning to the changing nature of the market. That's it.
Hmm... Let's take a channel breakout system with fixed TP-SL. So we must have a channel period, size of TP and size of SL. Three parameters already.
Yes, we can make the TP-SL self-adjustable, however, in these very self-adjustments - in themselves we get a bunch of hidden parameters, and often more than a direct indication of stop sizes.
You write it all right, Georges.
I've been through it all, all those billions of settings. In the end, according to my research, the key parameter is the observation time window or period, as you say. The rest is all nonsense and should be set once and for all.
I went through all this, all these billions of settings. In the end, according to my research, the key parameter is the observation time window or period, as you say. The rest is all nonsense and should be set once and for all.
When will you say goodbye to this nonsense? The time window is only one parameter, and far from being the main one - no more than the average temperature in the hospital. You trade on the current state of the market, not on the statistics on the time frame.
When will you say goodbye to this nonsense? The time window is only one parameter, and far from being the main one - no more than the average temperature in a hospital. One trades on the current state of the market, not on the statistics on the time frame.
i.e. a window of 2 candles
and we still have an interval, no matter how you turn it
the same is true for one candle, the interval is a timeframe
we have an interval, we have a lag, so we have a certain probability of losing
ha ha
Well, no.
Here, we take a system with no explicit parameters - every day at the same hour we look at what the previous bar was and open exactly one hour in the same direction. Such a system has very few degrees of freedom. There is no way to "tweak" it.
Now we begin to add parameters to it - for example, we open not for one hour, but for a certain time, set by the parameter. Plus we will set TP and SL, which are also set by parameters. Plus - we will not look at one bar, but five bars, and enter depending on this five-bar pattern. In addition, we will set a filter with some indicator, and not only one. As a result - we'll get a system with a hundred parameters, which can be "adjusted" for any symbol, any timeframe, for a long period of time... But will it work? I'm more than confident that it won't - precisely because it has many degrees of freedom - each parameter will be fitted in the optimizer based on history, and it will work on history. In real trading it won't.
I like your approach)
I would only like to add that you may fight fitting by two methods. The first one you have indicated - reduction of conditions.
But there is another method - its antipode.
Since you cannot get rid of the fitting, because any entry/exit point, SL, TP is all elements of the fitting, you must increase the density of the fitting on the sample so that it would maximally cover the sample and show its true colors.)
But this method can not be implemented on all algorithms. It is especially impossible to implement it in indicator algorithms, because they set narrow entry and exit points, limit the number of events and it is impossible to reproduce them on a sample.
But some nonindicator algorithms, based on probability, easily allow themselves to multiply...
For example, here is a fitting on the sample of 8 years, based on some probability algorithm. But there are only 585 trades in it.
Let's use the same algorithm and increase the number of deals by a factor of 10 so as to cover the sample.
The result is 5700 deals.
If the algorithm is useful and the system survives, then it is more stable than the first limited fit.
This happens as follows.
Since it is a probabilistic system and we are not bound by signals to entry and exit points, we enter whenever we want.
We divide the algorithm into independent scenario layers so that each next algorithm opens with an offset from the previous one
and run them in layers. Each one lives its own life.
But even this system will not survive a new story, it may be more stable, but no more than that.
The probability that in the new story, even at some small segment, there will be a coincidence with a sample of history tends to zero ...
Here is my previous development test of 13 months. EURJPY
Put it on real a month ago, so far the flight is normal. I have a rather large (emergency) stop, of the order, usually = maximal drawdown shown during testing + 20%. If a stop loss occurs, then I stop the work, find out the reason and fine-tune the algorithm or adjust the parameters, so that this section in the tester is tested normally.
Yuri, you are at your best with your off-the-wall profitability and beautiful pictures.
Reveal the mystery of the entry/exit conditions ... :-)
Yuri - as always you are on top of your game with off the charts profitability and pretty pictures.
Could you reveal the mystery of the entry/exit conditions... :-)
Replyed in person.
Answered in person.
One of the possible options for a VERY PREVIOUS GRAIL:
1. opening a position - on the reversal of ALL indicators...
2. Closing the position - by the reversal of the fastest indicator...
P.S. The Zig-Zag in the analysis - only as an ORIENTER for other indicators...
Well, no.
Here, we take a system with no explicit parameters - every day at the same hour we look at what the previous bar was and open exactly one hour in the same direction. Such a system has very few degrees of freedom. There is no way to "tweak" it.
Now we start adding parameters to it - let's say we open not for an hour, but for some time set by the parameter. Plus - we will put TP and SL, which are also set by parameters. Plus - we will not look at one bar, but five bars, and enter depending on this five-bar pattern. In addition, we will set a filter with some indicator, and not only one. As a result - we'll get a system with a hundred parameters, which can be "adjusted" for any symbol, any timeframe, for a long period of time... But will it work? I'm more than confident that it won't - precisely because it has many degrees of freedom - each parameter will be fitted in the optimizer based on history, and it will work on history. In real trading it won't.
I suggest not to get into definitions. You have one understanding, I have another. It does not change the essence. The definitions are simply different.
-----------------------
Here, we take a system with no explicit parameters - every day at the same hour we see what the previous bar was and open exactly one hour in the same direction. Such a system has a LOT of degrees of freedom. There is no way to "tweak" it.
Now we start adding parameters to it - say, we open not for one hour, but for some time specified by a parameter. Plus, we will set TA and SL, which are also set by parameters. Plus - we will not look at one bar, but five bars, and enter depending on this five-bar pattern. In addition, we will set a filter with some indicator, and not only one. As a result - we'll get a system with a hundred parameters, which can be "adjusted" for any symbol, any timeframe, for a long period of time... But will it work? I'm more than confident that it won't - precisely because it has a small degrees of freedom - each parameter will be fitted in the optimizer, based on history, and it will work on history. In real trading - no - precisely because it has LESS degrees of freedom and can easily be SHOCKED by fitting it to history.
-------------------------------