Repetitive patterns and other patterns - page 30
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
iModify:
Is $50 okay with you?
I don't know... Okay, okay. Half a grand and I'll stop reminding you of the eternal facepalm. It'll prove to everyone that you're responsible for your words and you know how to correct mistakes.
It's hardly an organised system, in the sense that some uncles are paying some suckers to troll. I think it is self-organized, everyone consciously or not understands that the more people who are misled, the more profitable they are. I don't know why they do it and they cannot do it properly, I will ask them: first they omit Martingale, then they say that all expert advisors are bullshit, then they proclaim signals, profitable PAMM accounts and asset management in general, and finish with saying that the system trading is a myth and the market is not predictable by TA and FA.
I feel that soon I myself will become like this, but my conscience is still struggling. But it is inevitable.
So do not pay attention. The dog barks but the caravan moves:) While you are on the right track, do not let yourself be broken by these provocateurs.
)))))))) Not soon, but we already have! But so far you go too far, too contrasting statements that look like mockery, people do not swallow.
How nice...
I propose to redirect the conversation from the flood, to a contest of formulas of market inefficiencies, "purely", that is, without regard to costs.
The conditions are as follows: We only look for patterns in the series, summarize the constant multiplier without applying any MM scaling. TS in the form of an indicator producing signals (1, 0, -1) and the indicator-tester, which calculates the TS-indicator.
Someone has to start first, at least with a simple "formula" or pattern that has statistical advantage, otherwise nothing will happen except the flood about the red "curves" and bases.
For example, what will your "indicator-tester" show, if you feed it with MACD signals?
C(t)= (EMA(p(t),P1)-EMA(p(t),P2))-SMA((EMA(p(t),P1)-EMA(p(t),P2),P3)
S(t)=bool(C(t)-bool(C(t-1));
I think we'd better calculate, not for some parameter(s), but for the MO of the functional by all principal degrees of freedom, on a representative sample (>1000 orders). Then it will be possible to get a reliable estimate of inefficiency of some "formula".
I don't know... OK, OK. Half a year of credits and I'll stop reminding you of the eternal facepalm. It'll prove to everyone that you're responsible for your words and you can correct mistakes.
What a beauty...
I propose to redirect the conversation from flood, to a contest of market inefficiencies formulas, "purely on the money", i.e. without cost.
The conditions are as follows: We look only for patterns in the series, summarize the constant multiplier without applying any MM scaling. TS in the form of an indicator giving out signals (1,0,-1) and indicator-tester which calculates TS-indicator.
I think we should make the testing indicator publicly available in order to standardize calculations - you never know how it will be assembled ... And different users may have different calculations for one algorithm that will lead to confusion.
That is, of course, if it comes to it at all.
iModify:
...
PS Patterns in the market exist. My aim is to finish my economic model and attach a pattern-based market model to it and it will be quite beautiful. You could start a public hedge fund. Dreams, dreams ...
Stop bullshitting)))) What a load of crap.
I've noticed a tendency in this forum, as soon as an interesting post appears, there are certain people who work off their salary, I don't know how much "10 cents" or "$ 1" per post.
I understand if *fuckers have written something useful, otherwise it's all the same- empty nonsense from a set of standard turns in their vocabulary.
It's understandable that it's a lot of bullshit.
But the fact that patterns (market inefficiencies) exist, they can be identified and you can make money from them, i.e. gpwr does not lie, is easy to see even with neural networks. The other compote is that a neural network is a black box, and therefore they do not provide useful, i.e. primitively human-readable information (like, the price has crossed the mach, so you have to buy or sell). And in principle there is no need in a human-readable form if a neuronet is embedded into the algorithm of an Expert Advisor and trades according to the patterns found, because in this case it is more convenient for a human to be guided by the trading system indicators on the forward testing.
If you do not believe in the existence of patterns or the possibility to detect them, you can easily verify them with the help of my free Expert Advisors in the Strategy Tester, on demo or even on the real account.
The patterns do exist, you can earn on them. But we should only take into account that these same patterns do not live forever, i.e. after some time after a successful forward they stop working. To make sure of it, you should leave a piece of history after a forward (about the same time as a forward) and watch how long a pattern lasts, to see how often you need to reoptimize it.
gunia:
I think we better calculate not for some parameter(s) but for the MO of the functional by all main degrees of freedom on a representative sample (>1000 orders). Then it would be possible to get a reliable estimate of the inefficiency of some "formula".
I don't care if there are a million deals in a sample. Any fool will be able to adapt the TS to historical data. The Graal for tester that lets out losses on forwards has existed for a long time. For example: Graal for tester.
The effectiveness of TS should be checked out of sample, i.e. on forward tests and, at maximum, to observe how long they live after forward passes.
I don't care if there are a million deals in the sample. Any fool can adjust TS according to the historical data. The Graal for tester that losses forwards already exist. For example: Graal for tester.
The effectiveness of TS should be checked out of sample, i.e. on forward tests and, at maximum, to observe how long they live after forward passes.
It depends on what kind of pattern and how many degrees of freedom the algorithm that recognises it has. For example, take a simple MACD, 3 parameters (periods). It can be fitted to a small portion of history by calculating the extremum of the functional by parameters. But what will happen when searching for the functional not for the maximum amount, but for the MI? For example, by running through all 3 parameters from 1 to 1000(p1>=p3>2*p2)? That would be just exploratory testing of the pattern. There is no fitting at all, hence back-end is unimportant as there is no optimization, we don't select the best curves on tests but calculate their average.
This should be done on a normalized series combined from different FI to get a more convincing picture.
It depends on what kind of pattern and how many degrees of freedom the algorithm that recognises it has. For example, take a simple MACD, 3 parameters (periods). It can be fitted to a small portion of history by calculating the extremum of the functional by parameters. But what will happen when searching for the functional not for the maximum amount, but for the MI? For example, by running through all 3 parameters from 1 to 1000(p1>=p3>2*p2)? That would be just exploratory testing of the pattern. There is no fitting at all, hence back-end is unimportant as there is no optimization, we don't select the best curves on tests but calculate their average.
This has to be done on a normalised series combined from different FIs to get a more convincing picture.
I purposely provided a link to a grail with few input parameters, i.e. one only needs to select the size of take and stop. The screenshot shows that the Expert Advisor has performed about 1000 trades on the history. But this grail will lose on spread on forward.
So the incorrect excuses like forward is not necessary, we need plenty of history deals, few input parameters and other nonsense are an obvious fallacy. Create a grail to fit the historical data, like two fingers on the pavement. Forward testing is the minimum tool to distinguish a tester's grail from the TS capable of detecting patterns in the form of patterns so that one is not surprised later by a loss.
I purposely gave a link to a grail where the input parameters are few, i.e. you only need to select the size of the take and the stop. The screenshot shows that the Expert Advisor has performed about 1000 trades on the history. But this grail will lose on spread on forward.
So the rotten excuses like forward is not necessary, but you need a lot of trades on history, few input parameters and other nonsense - this is an obvious fallacy. Creating a grail to fit the historical data is like two fingers on the pavement. Forward testing is the minimal means to distinguish a tester's grail from the TS capable of detecting patterns, so that you won't be surprised to see it plummet later.
Well you've got wizards there too. The sum of 3 differences of different timeframes. The usual trend principle. Matching TP\SL is lyric.
Here is the formula #2 from Reshetov:
C(t,P)=(MA(P1)-MA(P2))+(MA(P3)-MA(P4))+...+(MA(PN)-MA(PN+1))
S(t)= bool(C(t)-bool(C(t-1);
I purposely provided a link to a grail where there are few input parameters, i.e. all you need to do there is pick up the size of the take and stop.
There are a LOT of implicit input parameters - CVP history, to put it crudely
So some rotten excuses like forward is not necessary, we need a lot of trades on history, few input parameters and other nonsense - this is an obvious delusion. The problem is to create a grail, adapted to the historical data, like two fingers on the pavement.
You are mistaken. Creating a grail on history is difficult:
Forum on trading, automated trading systems and testing trading strategies
"Genetic algorithms are easy!" - to be continued?
2013.06.20 19:25
You take a piece of history you like. On it selects the intervals you like, for which the following will be summarized. For example, we take the last month and further analyze only the night intervals.
The task is to decompose the data and develop TS for this part of history (or more exactly for the selected intervals within it) to make TS show the maximal profit on it.
I.e. the task comes down to detecting the regularities for the maximum profit on the known piece of history.
It is evident that the calculation of the maximum possible profit on a piece of history is very simple. But we need to create such a TS in which the value of Profit / Func(AmountIN) would be maximal, where AmountIN is the amount of explicit and implicit input parameters of the TS, and Func is a function (the simplest linear function to start investigating: Func(X) = X).
For example, some people like some wonderful flat piece that these people consider to be almost a classical market flat (typical) state. People lay out this piece publicly and arrange a kind of competition for creating an optimal TS for this piece. Then the characteristics of the received TC and the ideas put into them are compared. And based on this analysis some understanding of formalization of flatness and more general notions comes.
P.S. It is written very primitively, but the basis of such manipulations allows us to grasp the essence of approaches to studies more complicated as well.
Here's another ZeroLAG MACD formula from the excellent Andrey Voytenko:
ZeroLAG MACD(i) = (2*EMA(Close, FP, i) - EMA(EMA(Close, FP, i), FP, i)) - (2*EMA(Close, SP, i) - EMA(EMA(Close, SP, i), SP, i)) ;
ZeroLAG MACD Signal(i) = 2*EMA( ZeroLAG MACD(i), SigP, i) - EMA(EMA( ZeroLAG MACD(i), SigP, i), SigP, i);
It will be interesting to see what advantage it has over a normal one. No lag indeed.
It is interesting to see what advantage it has over the average.
Well, it's very simple. Plot the BP of the price differences and the measures of the average. Then plot the probability (number of repetitions) distribution of values of this BP.
Who will have a higher RMS and thinner tails - that is better for trading.
P.S. The main thing is not to be stupid and not to investigate the goodness of the average around the clock. Because one average can be great for the day time, and another one - for the night time. In general, you can classify the averages in the same way.