To follow up - page 31
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
Это здорово смахивает на линии баланса и эквити. А что это на самом деле?
gentlemen, you are scary - they are :) .... from the surgeon - a big personal "thank you" to him
But the question was on topic anyway - how to determine the context for trading.... personally, for me the position (similar to yours) of market research by an expert is preferable - after all, we have come to the same opinion in this point - we are interested in profit..... or not so????.... and if so, then with all the complexity in coding there is a second possibility - let's say there are two instruments traded at different times and with different parameters.... who prevents you from making 4 of them? :)) .... in different "contexts"
..... in researching "just" the market there is a good chance of getting lost there, or even staying there forever.
in continuation of..... the author of the thread will allow????
So.... after one or two or three thousand optimizations, one remarkable pattern emerged - no one will ever be able to predict with a given probability the behaviour of the graph in the future - you can only believe it.... or disbelieve it, which is one way or the other.... and no "Prado...." will help here.... we are going to trade..... or aren't we?
for "follow up" ..... plagiarised....
I really liked the adaptation feature (from Peter) ..... so, all of a sudden, it turns out that the exp with random inputs works better than anything done with the finished trading idea...... what to believe?
This conclusion ("an exp with random inputs works better than anything done with a complete trading idea") definitely cannot be trusted. It's random.
This conclusion ("an exp with random inputs works better than anything done with a complete trading idea") definitely cannot be trusted. It is random.
Unfortunately, I was convinced more than once that an Expert Advisor with random inputs works better.
In parts of history, and even for quite a long time, perhaps. The problem is that we do not know in advance when they will meet and how long they will last.
Context
Context is a circumstance, a situation, a meaning that belongs to history and remains unchanged for some period of time (until it is leveled out) and that can be relied upon, kept in mind, pushed back for clarification. In general, some combination of external fundamental factors and previous market conditions.
A fundamental indicator comes out, and until the next one comes out, we work in the context of this indicator and combine it (the indicator) with the previous state of the market. A session or a season has started, we work in the context of the current session or season considering the history of the previous session. They don't change for some time, it allows us to concentrate on the current parameters, to analyze and determine the market condition and the nature of the price movement.
Market conditions
"under pressure", "bullish trend", "downside risk prevails", "at the opening of the European market trades with the rise", "market is frozen, waiting for the release of the indicator ...", gap.
Aspects, modifiers
"Key resistance line is located...", "resistance-support levels", "waiting for employment data".
Price movement pattern
"Bounce from resistance line", "price is in ascending channel", "triangle", "narrow (wide) flat", "near support", gap.
Re-read the whole thread. Wow, I don't remember that happening to me lately. Just an awesome thread. Ideological disputes between Weismanists and morganists almost do not count, these bourgeois are all the same anyway. There was some moment where I lost my bearings again (somewhere after page 20, when it came to FP). The three-dimensional picture put everything in its place.
There was a question: how does this whole clustering of context procedure differ from the usual optimization in the tester? Because even visually on the second image (top view) we're looking for a stable area of "warm" colour - just like on the optimization diagram in the tester (only colour is different there, it's green).
At first glance, there is a fundamental difference.
In the tester, colour stability areas are strictly bound to the trade algorithm parameters. We have found the area of colour stability -> restored the "right" limits of parameters of the TS making it stable, and thus, supposedly, we have optimized the TS.
Here what? Found the context colour stability region -> correctly clustered the FP -> restored the correct limits of the context filter parameters. So we have optimised the TC again! Context filter - that's what TC is all about!
What's the difference, colleagues?
P.S. I see another difference. In the tester, we do not see individual trades. We can only see the results of testing on the entire history. And it is them that we filter out. But in the tester, we can see individual deals that are not moving anywhere and filter them out.
Damn, I just had a hope to adapt the optimizer for this procedure. Now it's gone again :(
P.P.S. It seems that not all so grim. I just need to think about it.
вдруг, выяснилось, что эксп с рандомными входами работет лучше, чем все что было сделано с зконченной торговой идеей...... чему верить?
A truly random expert must have random entry moments, random entry direction and random exit moments. If at least one condition is violated, we cannot speak of complete randomness. If a truly random expert has a reliably proven statistical advantage, one has to assume that the market works purposefully against popular (standard) trading ideas. I wouldn't be surprised by that, complex systems always tend to compensate for external influences.
Context
...Market condition
...Going back to definition of terms? :) Many people here understand the context as the market condition.
Re-read the whole thread. Wow, I don't know if I've seen that lately. It's a great thread. Ideological debates between Weismanians and Morgans almost don't count.
Not Weismanians and Morganians, but Morgan and Flint himself... oh... Weisman :)
There was a question: how this whole procedure of clustering context differs from the usual optimization in the tester?
...
And what is the difference, colleagues?
In searching for clustering we work with fixed points in time, no matter if these are ideal inputs-outputs, a reference trading system or something else. The tester, on the other hand, will always give the same result with such fixing. So technically, the problem of using the optimizer reports comes, imho, to the problem of replacing the final balance (or other standard optimization criterion) with the needed value, at least within a normalizing constant. Since the price chart can be used as a GSCH :), there are some possibilities for manipulation. But it is incomparably easier, imho, to use a suitable program to display state parameters.
And the spectre of optimization and fitting will always appear to us :)
We should make it so that one optimization run opens only one trade with a specified number. Accordingly, the number of runs will be equal to the number of deals of the system (the trading system parameters will not be changed). In the external parameter of the Expert Advisor, we need to enter the number of trades and "optimize" by it. It seems to be possible to implement it, though not the first time.
But we also need to teach the optimiser to report context parameters. This is probably a question for either xeon or the developers.
Of course, we cannot force the optimizer to cluster the context. Although one can try to get out of it here as well :)