Help on Developing "The MOTHER of ALL" Trading System

 

Hello:

I would like to develop a system where an EA with DLL programming will login to a website via SOAP or some other protocol to retrieve trade signals in real-time. It will authenticate with an username/password combo and the EA will just buy/sell according to the signals. Obviously some of the basic functions will need to be programmed into the EA such as MoneyManagement, Trailing Stops, Trade Time (to avoid news trading volatility), and/or any other disaster prevention routines that in the event that the signal server goes offline your trade will still work out.

The second phase of this development will be to run as many profitable (65%~83% profitability EA = profitable) EA as possible under the sun in forward testing with the same TF, and relaying trade signals to a webserver. Then we will run a statistic probability on the webserver and depending on the RISK tolerance we will generate SELL or BUY signals. If we have a total of 10 EA running on all 4 Major Pairs, and 7 EA is generating BUY signal on GBP, we will obviously send a BUY Signal out. If we have conflicting signals, then we will not do anything. After a couple of months of testing we will have a result with statistical combinations of different EA confirming each others' signals, and by choosing the best combination based on statistics, we might actually have found the "HOLY GRAIL" of trading. I am just throwing in the most obvious example, but I think we can develop more complex statistical models once we get enough people to pitch in and brainstorm.

This idea is hypothetically sound, and infrastructurely feasible, and absolutely tempting. I have two webservers that run linux(fedora) and Windows 2003. I also have about 10 computers sitting in my office (P4 3.0Ghz machines) doing nothing. If someone is willing to set up this experiment, I am willing to provide 10 computers to run EA 24 hours a day for as long as needed, 2 dedicated webservers (on a COLO) that will support ASP.NET, PHP, MSSQL, or anything that you need to make this project possible. Please PM me if you have the skill to program EA, C++, and setup the WDSL on my webserver, I am ready and willing to work with you.

People that I think I will need for this project:

1. C++ Programmer to Program the DLL interface.

2. Metatrader EA Programmer for the clientside EA Function.

3. .NET programmer to program WDSL service

4. Statistical Analyst or someone who is really good with statistics.

5. Interface to (relay)send signals to the webserver from the 10 EA farm PC's.

6. Beta (Forward) Testers

7. Project Manager

Thank you and I look forward in hearing from you all.

neo1001

 

thats alot of work to essentially follow someone else's signals. you say youll buy when a buy signal is the majority signal but signals come at different times with different stops and different profits. good luck with this but i have my doubts about it all. sorry.

 

Go for it!!! Neo

Nothing ventured nothing gained. Using disimilar EA trading philosophies I think would be important because every EA has its strength and weakness. For example, if all were breakout EAs or hedging EAs or simple trend system EAs then you would simply compound the weakness of that system 10x over. Therefore, EAs generating trade signals based on a different philosophy AND different TFs would seem preferrable.

Sounds like a task you're up to, Neo. Go for it!!!

COTtrader

 

What do you mean with tradesignals? What I read in scientific literature is that one of the better predictive signals is "orderflow". If you know which big bids and offers are available, you can predict the moves of currencypairs better. When you build a neural network, it gets even to around 75% correct. So if that are the tradesignals you can read somewhere, I am interested.

However if you want to build the system mainly based on signals from EA's who could be profitable, I am less interested. In that case why not using those EA's alone?

BTW: I am qualified to do all points, except point 3.

 
trevman:
thats alot of work to essentially follow someone else's signals. you say youll buy when a buy signal is the majority signal but signals come at different times with different stops and different profits. good luck with this but i have my doubts about it all. sorry.

I agree with you on the fact that there are different stop loss/take profit for different EA's. But then again if we were to do this we have to be flexible with the concept and maybe we can have a One-Size-Fits-All SL/TP for all the trades. Like a 20 Pip TP for 50% of the position, then trail it with 20 PIP or something similar. Also to answer the 2nd part of your question on signals coming at different times then we can set up a minimum EA confirmation, say that we need at least 2 EA's with the same signal before we send out a trade signal, that way we are filtering out trades.

I mean the concept is not 100% yet, that is why this is so interesting. We get to brainstorm it and make it worthwhile.

 
COTtrader:
Nothing ventured nothing gained. Using disimilar EA trading philosophies I think would be important because every EA has its strength and weakness. For example, if all were breakout EAs or hedging EAs or simple trend system EAs then you would simply compound the weakness of that system 10x over. Therefore, EAs generating trade signals based on a different philosophy AND different TFs would seem preferrable.

Sounds like a task you're up to, Neo. Go for it!!!

COTtrader

Thank you and my thoughts exactly. We have to choose a different school of EA's based on different philosophy on the SAME TIME FRAME. Is like going to several doctors to get 2nd opinions. And if you have 3 or 4 doctors giving the same opinion on something, then it is probably right. Same concept here.

 
JosTheelen:
What do you mean with tradesignals? What I read in scientific literature is that one of the better predictive signals is "orderflow". If you know which big bids and offers are available, you can predict the moves of currencypairs better. When you build a neural network, it gets even to around 75% correct. So if that are the tradesignals you can read somewhere, I am interested.

However if you want to build the system mainly based on signals from EA's who could be profitable, I am less interested. In that case why not using those EA's alone?

BTW: I am qualified to do all points, except point 3.

Ok, this is not to advertise someone's tradesignals, but I know exactly a service that does what you described here. The beauty of this EA is that it could easily be adapted with that kind of trade signals as well. Heck, as a matter of fact, any trade signal that any company provides can be adapted into this EA, even manual trading system that some people in this forum who swears that "is too complicated to be written as an EA" can be adapted into this EA. The foundation of this EA will handle all that infrastructure. Let me break it down into 3 Parts:

1. Client Side EA: Handles retrieval of trades signals, placing orders, trailing stops, take profit, money management, profit protection, etc... Basic EA Stuff.

2. Signal Server: Receives all trade signals from "Different Sources" and makes a decistion to (1) Queue the signal for additional confirmation, or (2) Generate a signal based on the Confirmed Queue, (3) Dumps the Queue due to conflicting signals.

3. SignalSource: Handles sending trade signals to the signal server, could be a DLL that monitors all trades and send those trades to the signal server. Just imagine all of the people who say about a "MANUAL SYSTEM" that cannot be programmed into EA, now we can mimick their trades and send to the signal source.

Imagine that the infrastructure of this project will handle signals from different sources, not only EA's, creating an interface to many different platforms. And by tapping into hundreds of different trading systems (eh.. sorry admin for mentioning a different website, but collective2 has over 3000 different trading systems we can adapt) we can achieve what no one has ventured before....

 

if you have any EA that you have forward tested for a few months with good results, please kindly post them here. I am gathering EA's now for the signalsource part of the project.

Unfortunately, I don't think we can do any Hedging strategies, that is possible for this project but I think at this stage it might complicate things. We have to create different set of rules on the client EA to work with Hedging, and the webserver needs to send the signal out together, and the signalsource might have to be tweaked as well. So no hedging for now...

 
neo1001:
Ok, this is not to advertise someone's tradesignals, but I know exactly a service that does what you described here. The beauty of this EA is that it could easily be adapted with that kind of trade signals as well.

Can you say which service you mean?

I am not sure about your system. If I understand what you want to do is collecting all kinds of tradesignals and somehow make one tradesignal from that. My first question is, how would you do that? Take just the majority? Don't look at systems that did bad in the past? Or weigh each system with their results in the past?

My second question is, will that tradesignal really be better? If you take a lot of tradesignals, how do you know that the combination gives a better signal. Indeed 3 or 4 doctors will give a better opinion than 1 doctor. But I am not convinced that tradesystems do the same.

 
JosTheelen:
Can you say which service you mean?

I am not sure about your system. If I understand what you want to do is collecting all kinds of tradesignals and somehow make one tradesignal from that. My first question is, how would you do that? Take just the majority? Don't look at systems that did bad in the past? Or weigh each system with their results in the past?

My second question is, will that tradesignal really be better? If you take a lot of tradesignals, how do you know that the combination gives a better signal. Indeed 3 or 4 doctors will give a better opinion than 1 doctor. But I am not convinced that tradesystems do the same.

That is why "Statistics" will come in pretty handy. We will start by taking the majority of signals, then we will narrow it down with combination of signals, then we will run different statistic models and determine which signals will work together and give near 100% and which signals when conflicting will produce random results. By having all of that information, we will be able to create a pretty powerful and robust neuro-net type of system. I think many people say that different EA works off different market conditions, and by putting them to work side by side, we can see their results in a better light. Also we can try to assign a different weight to EAs depending on past performance, or just do it automatically without bias using a counting routine.

The service which I was referring to was from Rob Grespi with Dmitri/Felix at kingforexsignal. I used to be a subscriber and Rob has a special link with his broker and he can see Bids and Orders piling up to go one direction or another. Pretty much what your broker sees when their trading-desk take a position against you when you are on the other side. So he will say I see orders piling up to go long, then he will issue a long signal. But he is not 100% because sometimes the market will reverse and you will lose 10~12 pips. But he is about 80~85% accurate if you can follow his signals and trade without hesitation. If you have issues with closing losing trade, then you might have problem with his methods. Sometimes he will close a trade because he didnt like the market condition, but then eventually the market goes with his original direction. He will take both long and short positions in a trending market, and there is one day that I remembered that he took about 10+ trades in a London session. So you cannot be light-hearted to trade his methods, if you hesitate to follow, which was my problem, you might fall on the wrong side of the market.

 

I know Robs service, but it is too expensive for me, especially because I can only trade 2 days each week. And you recieve only advices of Rob, not the original signals. I am mainly interested in the original signals.

My suggestion is not to build your original setup but a more simple one, for a start. You can take the EA's from which you want to combine the signals. Those EA's should be adapted so that instead of trading they send their tradesignals to some file. Then the "main" EA will read that file, do the statisticsand calculations and then combine those tradesignals to one signal. That can be tested on a demo, so you can see if the combination-signal is really good. I don't believe however that it can be 100% right.