The perfect mechanical trading system. - page 9
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
Shit... I really don't like the way the MT4 optimiser works. It seems to show only profitable passes in the report. And with what "density" this or that profitable pass is surrounded by unprofitable ones, it remains unclear... And this is very important.
BTW, this problem also plagues me. Maybe in the 201 release of Metatrader we will see? Who to write a suggestion to there? :)
xeon, it's not going well with the month. I'm trying to optimise for a month on M5, but so far I've had a maximum of 12 trades. This is too little. My Expert Advisor refused to trade on M1 for some reason. In short, I know I will have to write my own optimizer.
I think the problem is not the optimizer but the Expert Advisor - MACD Sample is not intended for small timeframes
TakeProfit=120
TrailingStop=30
MACDOpenLevel=0
MACDCloseLevel=100
MATrendPeriod=26
FastEMA=12
SlowEMA=4
SignalSMA=6
I optimized it on M5 from 2006.09.01 to 2006.10.01. 12 trades were performed, which is quite strange considering the trading at M5 for a month. The profit was 2385.63 AZN and the drawdown was 0. Having used these parameters in my trading robot, I obtained 463 deals (much more real though IMHO it is too much) 148 of them were profitable and 7224.34 portraits of the president was a net loss. Some kind of mysticism...
Firstly, I have not so complex robot that took more than two hours to compute. That's why we are using mql for now, and okay, but for serious work I suggest to give up mql and write the TOTAL trading robot in C. We will have to write both the optimizer and the strategy itself. There will be a lot of counters there, and we would like to get a result at least within a couple of weekends. So we should not neglect the speedup of 20 times given by C. After all, the strategy will be called by the optimizer in a loop, many times. Secondly, I think I understand the optimization problem itself. Mathematicians, pay attention! Listen! Those who will tell me how to do it in a couple of weekends will drive a Rolls-Royce, and don't care about the 600-margin like a cheap proletarian kopeck :)))))))))) So, there is a function of many variables (I explain for the obtuse, it is our strategy). It is required to find its extremum. But NOT ANY extremum, but an extremum smooth enough. The extremum should be smooth enough so that there are no bad points in the vicinity of the "good" point. It should resemble the bottom of a concave bowl, like a paraboloid of rotation, rather than a forest of columns sticking out of a gently descending floor. If this extremum is global, so much the better. If not, well, you can't kiss all the girls and make all the money. The main condition is smoothness, not globality. Without smoothness it will only be interesting for history trading. I think maybe Monte Carlo is more promising than genetics here, for such an extremum should have a rather broad attractor ? One more thing. Perhaps a new such extremum should be identified in the vicinity of the old one, because everything is smooth and we assume that the market is also quite smooth on a daily scale. Who has an opinion about this ?
Shit... I really don't like the way the MT4 optimiser works. It seems to show only profitable passes in the report. And with what "density" this or that profitable pass is surrounded by unprofitable ones, it remains unclear... And this is very important.
So, what I want to say about my plays with the optimizer. First, my robot is not very complex, so it took me more than two hours to calculate it. That's why we are still using mql, but for serious work I suggest to give up mql and write the whole robot in C. We will have to write both the optimizer and the strategy itself. There will be a lot of counters there, and we would like to get a result at least within a couple of weekends. So we should not neglect the speedup of 20 times given by C. After all, the strategy will be called by the optimizer in a loop, many times. Secondly, I think I understand the optimization problem itself. Mathematicians, pay attention! Listen! Those who will tell me how to do it in a couple of weekends will drive a Rolls-Royce, and don't care about the 600-margin like a cheap proletarian kopeck :)))))))))) So, there is a function of many variables (I explain for the obtuse, it is our strategy). It is required to find its extremum. But NOT ANY extremum, but an extremum smooth enough. The extremum should be smooth enough so that there are no bad points in the vicinity of the "good" point. It should resemble the bottom of a concave bowl, like a paraboloid of rotation, rather than a forest of columns sticking out of a gently descending floor. If this extremum is global, so much the better. If not, well, you can't kiss all the girls and make all the money. The main condition is smoothness, not globality. Without smoothness it will only be interesting for history trading. I think maybe Monte Carlo is more promising than genetics here, for such an extremum should have a rather broad attractor ? One more thing. Perhaps a new such extremum should be identified in the vicinity of the old one, because everything is smooth and we assume that the market is also quite smooth on a daily scale. Who has an opinion about this ?
Some kinds of neural networks are in theory very good at this task. Please contact njel. He seems to be an expert in such matters.
favoritex, if you don't mind, please send me the link where you can read more about it. Or post something about it. You don't want to reinvent the wheel yourself... By the way, the 2% error is not that big, if you don't set exact targets, but use trawl or short take and big lot. What is more important is the 70% directional error...
I haven't found anything good on this subject on the web. I'm studying with the book "Applied Data and Knowledge Analysis Methods". Tomorrow I'll try to take a picture and post the LGAP chapter.So, there is a function of many variables (for dumbasses, this is our strategy). We need to find its extremum. But NOT ANY extremum, but an extremum smooth enough. The extremum should be smooth enough so that there are no bad points in the vicinity of the "good" point. It should resemble the bottom of a concave bowl, like a paraboloid of rotation, rather than a forest of columns sticking out of a gently descending floor. If this extremum is global, so much the better. If not, well, you can't kiss all the girls or make all the money. The main condition is not globality, but smoothness. Without smoothness it will only be interesting for history trading. I think maybe Monte Carlo is more promising than genetics here, for such an extremum should have a rather broad attractor ? One more thing. Perhaps a new such extremum should be identified in the vicinity of the old one, because everything is smooth and we assume that the market is also quite smooth on a daily scale. Who has an opinion about this ?
In relation to the above opus some questions arise:
1. What does an attractor have to do with an extremum (local or global) in
What does an attractor have to do with an extremum (either local or global) in n-dimensional optimization space?
2. What is an "attractor width"?
Besides, in his remarks the author implicitly suggests that the optimization space at the fitting interval
interval, in this case a week, will be static or nearly static.
On what is this assumption based? If the optimization space at this
interval is not static after all, then how can we evaluate its dynamic
How can we estimate its dynamic characteristics, i.e. how it changes over time?