Shall we discuss an expert based on coding?
In principle, the expert has already been implemented, but I would like to discuss the idea and the model, and would like to hear criticism and weaknesses of the model.
So here is the gist:
...
Interesting method, as I understand it, the analysis, the collection of codes, for a certain period is done every day. That is actually you have carried out a forward test for 1.5 years, optimizing (picking up codes) every day and get a positive result. Or the test was carried out in some other way? Try to run forward test for 5-10 years, the more the better, and you will see the complete picture. The last ~3 years it's pretty easy to pass the forward test with a portfolio of strategies or instruments.
the collection of codes for statistics took place once in 3 years, there is no more reliable quotes history, the codes are recorded in the database and further there is just scanning for deterioration of statistics on the arrival of new codes
the method is interesting as we trade not 1 or 2 or even 3 events in the indicator but it is a sort of diversification, we cannot decrease the statistics for all codes (events) at once but the codes will be deleted (replaced) as the probability gets worse
Moreover, we select only those codes where the stop is not more than 2 volatility channels, i.e. 30-40 pips
...
Besides we select only those codes where the stop is not more than 2 volatility channels i.e. 30-40 pips
- www.mql5.com
Your method is quite powerful. It is difficult to add more to it. You can even say that you are choosing 100 strategies out of 2000 000 ones and joining them into one system and statistics for all of them. I am moving in the same direction, but not to the same extent. If there is more room (relative to resources), in addition to forward testing, you can also include the system of increase and decrease of positions. That is, you have the criterion that you mentioned above, and between this range, you can gradually reduce the amount of position/subposition. Coefficients can be picked up during optimisation.
Our MM is based only on mathematics of money management, everything is precisely calculated, but the total number of lots can of course be divided into several entries, we will implement an additional entry within the total share of the limit order, i.e. at the best price with the same stop and profit
Regarding increasing or decreasing the position volume, this only applies if in the signal and position conditions another signal with a different probability and code comes. If the probability is higher the position can be increased by the portion corresponding to the new probability and the profit factor, but if the signal is weaker, the portion (lots) can be closed or even reversed.
Your method is quite powerful.
By the way, the method showed that many things that seem to the eye as a super input to the statistics are grey with 48%/52% probability and vice versa places that I would not even pay attention to have 80% probability and higher
i have also tried several indicators that are also super according to feedbacks and they also give only gray statistical data.
Then this indicator runs through the entire history on startup and looks at what happened to the price after each code...
The statistics starts with 100 values)
I.e. it is desirable to find the result of the signal code in the history about 100 times...
Statistics start at 100 values)
i.e. it is desirable to find the result of the signal code in the history of about 100 times...
i agree, not all the codes have statistics 100 times, but there are those that are 10 times and always 100%, can you get past that? we just believe that the next one will not come in our favour. it's not 100% but about 90% chance. it means we can still trade, but the remark is correct, i absolutely agree that this is a weak side, but we do not have enough reliable history to see the big story
imho a big story isn't very good either. Prices at the time of Christ are unlikely to add anything useful to the data)
But the less such cases in history, the more trading is based on random inputs rather than on statistics, although it may be more profitable)
I suspect that the more times the code is in the history, the closer the result is to 50/50.
We did the statistics for two intervals of about 1.5 years, in both intervals the statistics are stable, i.e. the result is comparable and does not swim to the worse
i.e. one signal is tested on two parts of the history, and afterwards it is checked if the result can be trusted?
Move to a lower TF. There, the number of bars over three years is enough for any statistic. For example, M5 is about 75000 bars for 2011.
I suspect that the more times a code is present in the history, the closer the result is to 50/50.
I.e. one signal is tested on two parts of the history and then the result is checked to see if it can be trusted?
It is clear that the result depends on the number of codes in the history, but the point is that there are many codes with a probability of 75, and there are few with a 50/50 result, so there is a lot, but "sultanas" are found, so we decided to develop the topic
each code is checked for 1,5 years, for example 75% probability, then for other 1,5 years - 65% probability, there is no controversy, it means that there is a possibility that in the next 1,5 years we will stay within the 60% probability, besides some codes give probability not worse with the profit factor 1.5 - and this means that even 50/50 can give profit, in any case you should constantly watch for changes in the statistics for the codes
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
In principle, the expert has already been implemented, but I would like to discuss the idea and the model and hear criticism and weaknesses of the model.
So here is the gist:
we take an indicator (my own in this case) that in principle looks like IASD, analyze its directions, extrema, crossings 0, signal extremums, signal extremums from the higher frame, in short, many things and following the algorithm we assign a 4-digit code to every bar (in total we have about 2000 different codes) depending on the indicator movement, an analog of patterns on the indicator only, we write it all into a buffer.
then this indicator during the start runs the entire history and looks what happens to the price after each code, for example, what has reached the upper level-target or lower and writes it all down into the file, the resulting file is studied in excel and look for codes, after which the probability of winning is above 65% and build a file with the codes found (we had about 100 codes on average 1 code per day) which then trades expert (on m15), that is, if the indicator shows the code, which is in the database, then trade it in the appropriate direction, its profit factor and its probability
As for money management, we trade the optimal share calculated on the basis of the profitability factor and the probability of its execution, taking into account the fact that we assume that the next trade will not be in our favour.
average success probability for each asset is 70%, the profit factor is from 1 to 1.5 per trade, there are about 100 signals with the probability not lower than 65% for each asset
the statistics was done in two intervals of about 1.5 years, in both intervals the statistics is stable, i.e. the result is comparable and does not swim to the worse
I would like to know what else to take into account to reduce the possible risk.
I would like to discuss, maybe someone would also like to do something like this, so far we are delighted with the results, but experience tells us that somewhere there are pitfalls, but where?