Trader's self-deception: distrust of forwards. - page 4
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
There is an EA. It has 10 settings options - 10 sets. Optimization is the essence of choosing one of the ten sets. Right?
No, it is not.
1. An Expert Advisor with 10 variables can have an infinite number of setting options - depending on the step for the frames, which you have chosen. The minimum number of options for selecting (optimizing) one variable is 2. But even then, with 10 variables, the number of configuration options would be 1024. Accordingly, by storing these variants of settings as SET files, you get 1024 variants of settings.
And by selecting a set you like, which is obtained "On the whole history" you will not get any forward. You just get another backtest both left and right. The exception is the starting point of your segment, if it has an unoptimised history on its left - but this is a case of backward forward, its disadvantages we've already discussed.
There is no such thing as a reverse forward. It doesn't work and it shouldn't.
You don't get it. It doesn't matter how many sets there are. It doesn't matter if it's 100500.
I propose optimization optimization: you are going to run 100500 tests many times from different dates, but over a short period (e.g. half a year), and close the whole history in some step (e.g. month). I.e. the same data will be recalculated many times. And this can be done once over the entire history, and analyze the ready array.
And the set should not be liked on the whole history, but on a specified interval before the checkpoint.
In other words, the running of each set can be regarded as a certain financial instrument that can be traded only long. And the task is to look through the recent history at any point of time (the period of "optimization") to choose the only set, which is most likely to grow to the right of the check point.
However, my business is strategy)))
If you want, reinvent the wheel.
Propose the best method of analysis.
You don't get it. It doesn't matter how many sets there are. It can be 100500.
And this can be done once over the entire history, and analyse the ready array.
And the set must be liked not on the whole history, but on the set interval before the checkpoint.
In other words, the running of each set can be regarded as a certain financial instrument that can be traded only long. And the task is to look through the recent history at any point of time (the "optimization" period) to choose the single set that is the most likely to grow to the right of the check point.
Optimisation is a choice of settings, "Once" will not work, simply because there is more than one set. If we choose a combination of settings along some stretch, then whichever stretch within it we take is not a forward by the definition of a forward, so there is simply no subject for discussion. A "checkpoint" may only be the end (or, at the very least, the beginning) of an optimized segment. A forward starts where we have finished optimising. And the "probability that it will grow to the right" is best estimated from the set of such "right" segments that have not been optimised at the time of checking. You apparently have no idea of the role of the forward. A forward is such a time machine for your adviser. We cannot see into the future, but we can model the future for your EA. In your scheme, you show the adviser his result in the future, where he has already been, and say: "Look up to this place.
Optimise, drive a test on the optimisation result you like. Choose the worst period of the test run and test only that period. If there is profit or at least zero profit there too, you can selectively test a few more periods out of optimization. And only if all the tests are positive, you can try on the demo.
How about this.
1.Take the current week or month of July. Optimise the owl.
2. Run it on the best parameters from the beginning of the year. Determine the week or month with the best result. For example March is the best.
3. Next month is April. If "inertia of profitability" has captured April as well, then set it for the demo in August.
How about this.
1.Take the current week or month of July. Optimise the owl.
2. Run it on the best parameters since the beginning of the year. Determine the week or month with the best result. For example March is the best.
3. Next month is April. If "inertia of profitability" has captured April as well, then set it for the demo in August.
"Profitability inertia" is when, ideally, every forward is profitable. You never know what will happen in August - analogous to April or December. That the same setups can be profitable in different periods is natural. Only few have studied these waves. And no one has studied the forward waves. Theoretical analysis by months in each case may suggest what factors influenced the profitability in different periods. I have a couple of ideas to try and catch these specific cycles, but so far all my experiments in this area have shown that the usual sequential, rather than selective analysis works better. The only thing I've found so far is the approximate duration of correlation clarity for some instrument pairs. Something like 1.5 -2 years.