The inverse method of programming an EA.

New comment
 

I've been thinking for several days an idea for which I can't find an answer.

I was wondering how could I develop an EA in a different way than before. I will explain my idea with one example

I know the past prices of EURUSD of the last, say, 3 years. I want to develop a strategy that, knowing this price series, do the best trades, with minimum drawdown and maximum benefit. Having these 'best trades' for this period, I want to analyze which patterns lead these trades to open or close, if there is a common pattern.

I think this is a problem of Artifficial Intelligence, or Machine Learning. And I haven't found yet a way to achieve this project.

Could someone help me out find a way of doing this?

 

You may want to Google harmonic pattern

 

yes thats the way i guess, analize 3 years of data,find the PATTERN, code a bit and then, become a billionaire, is that your plan?

 

good luck 

 
its something like taking a picture of crowd on a main street and then expect those same people to get togheter
 
@onlysolo:
No, it's not like that. It's like taking a photograph of a city population, and according to their income, their religion, their race, find which pattern makes them agrupate.
You can laugh if you want, buy if this is done in other fields of study, why not in forex?
 
@phi.nuts
Don't understand what you mean.
 

In any sense, this is the same problem of backtesting and datamining, but I do agree with onlysolo, you will get just a past picture, because you are overfitting this learning.

I do agree about your perception of Machine Learning in this case, too. Artificial Neuron Networks learning and training algorithms, for example, try do this.

But the big problem still is solve the "else" in the algorithm bellow.

As future_real_scenario variable is just a probability, I still believe that find algorithms and money management to achieve the highest probabilities (using AI or not), and using backtesting/forward testing, are the best way.  

if (best_past_market_scenario==future_real_market_scenario) {

        // apply the risky billionary plan trading pattern (inverse method)

}

else { 

        // use known trading methods 

}
 
figurelli:

In any sense, this is the same problem of backtesting and datamining, but I do agree with onlysolo, you will get just a past picture, because you are overfitting this learning.

I do agree about your perception of Machine Learning in this case, too. Artificial Neuron Networks learning and training algorithms, for example, try do this.

But the big problem still is solve the "else" in the algorithm bellow.

As future_real_scenario variable is just a probability, I still believe that find algorithms and money management to achieve the highest probabilities (using AI or not), and using backtesting/forward testing, are the best way.  

In this case, use of neural networks or genetic algorithms should be the way to go. However, just like figurelli stated - with such approach, you will be fitting your strategy to the past data. And this means that most possibly your strategy will eventually have great results on the testing period, but in walk-forward testing or live trading it will do much worse or even go bankrupt. For more info, search "curve-fitting backtesting" in google.

If I understand you correctly,  something similar to what you are planning to do would be using neural networks for recognizing candlestick patterns. However, still the biggest problem would be that probably almost never you will find exactly alike patterns, just something similar. And this means that in your strategy, you would have to code a mechanism to determine, whether something is similar to already known pattern (e.g. candlestick bodies or shadows differ only by an acceptable amount of pips) or not.

New comment
Reason: