Your opinion...EA?
- Thread starter PPatrick
- Start date
Pharaoh
Brigadier General
- Messages
- 20,308
If you have a mechanical trading system, one based on If this happens, then do that, it's usually possible to program it into an EA.
A super simple example: If a 10 EMA crosses above a 20 EMA, then go long. If it crosses below, then go short.
Naturally, your EA needs to be more complex than this. There needs to be a way to set the size of trades. You also need some sort of rules to exit the trade.
The issue is that your EA isn't human. An EA can't look at the charts and say "Wow, the market looks crazy today. I think I'd better stay out." If you don't build in something along the lines of "Do not trade if price changes more that X pips per minute", then it will happily open and close trades while the market goes into utter panic at some unexpected news event.
That being said, here would be my plan for developing one.
First, develop a very good technical indicator based system that allows almost no user discretion. Keep risks small (2% of account per trade or less). Backtest it to verify potential profitability. Forward test it on demo for a month or two.
You need to have the risk/reward ratio for each trade be good enough that when you compare it to your winning percentage, you come out ahead in the end. For example, if you risk $1 to make $1.01 or more, then a 50% or more winning percentage would make the EA hypothetically profitable. If you risk $1 to make $4.01, then the winning percentage needs to be at least 25%. naturally, you want to do at least a little better than averaging 1% gain on the amount risked for every 4 trades.
Turn it into an EA. This will make more backtesting and playing with settings a lot easier. Don't over-optimize it. If you perfectly fit it to 10 years of old data, it will fail on next month's new data. What you are looking for is to limit drawdown while keeping a reasonable winning percentage and profitability.
This is where most commercial EAs fail. They are over-optimized on old data, and the market doesn't repeat itself. If the optimization data used is very recent, they may work well for a little while. So far, all those "never lost a trade in 10 years" EAs end up mysteriously losing multiple trades as soon as they are sold.
The next thing to do is look carefully at your data from your backtests. Is there a certain time of day that the EA fails more often? Perhaps the strategy works best only during the London trading session. Maybe it works well except during the New York session. Find out. If you do see a consistent pattern like this, program the EA to not open trades during those times.
Does the EA consistently have issues on certain days of the week. Maybe it wins more on Fridays and loses more on Tuesdays. If so, lock out trading on the losing days.
Look at how it performs around major news releases. If it gets nailed by Non Farm Payrolls, tell it to stop trading for a few hours around that release. You could even put in a user selectable feature where the major news events for the upcoming week or month are entered to lock down trading at those times.
Build in a volatility indicator or two. Then test settings of that to determine minimum and maximum volatility for the EA to perform well.
Your EA will now trade a lot less, but will trade at the best days and the best times for the strategy you devised. It will also skip trading around news events that could upset the strategy and will avoid trading when price action is too slow or too fast.
Some EAs do not set a SL, and instead have a "hidden SL" to keep the broker from trying to hunt the stops. A hidden SL is a good thing, but you really need a SL to be set. This protects you if the EA gets turned off, the computer gets turned off, or the connection to the MT4 server is lost for any reason. If you do use a hidden SL, set a hard stop farther out. Put it far enough away to make it hard for the broker to spike you out, but close enough so that the account won't be seriously damaged if all open trades hit the SL at once.
Set up a dozen or so demo accounts and let it run with different settings. If it's still doing well after a month or two, you can think about opening a small live account. using the best settings.
A super simple example: If a 10 EMA crosses above a 20 EMA, then go long. If it crosses below, then go short.
Naturally, your EA needs to be more complex than this. There needs to be a way to set the size of trades. You also need some sort of rules to exit the trade.
The issue is that your EA isn't human. An EA can't look at the charts and say "Wow, the market looks crazy today. I think I'd better stay out." If you don't build in something along the lines of "Do not trade if price changes more that X pips per minute", then it will happily open and close trades while the market goes into utter panic at some unexpected news event.
That being said, here would be my plan for developing one.
First, develop a very good technical indicator based system that allows almost no user discretion. Keep risks small (2% of account per trade or less). Backtest it to verify potential profitability. Forward test it on demo for a month or two.
You need to have the risk/reward ratio for each trade be good enough that when you compare it to your winning percentage, you come out ahead in the end. For example, if you risk $1 to make $1.01 or more, then a 50% or more winning percentage would make the EA hypothetically profitable. If you risk $1 to make $4.01, then the winning percentage needs to be at least 25%. naturally, you want to do at least a little better than averaging 1% gain on the amount risked for every 4 trades.
Turn it into an EA. This will make more backtesting and playing with settings a lot easier. Don't over-optimize it. If you perfectly fit it to 10 years of old data, it will fail on next month's new data. What you are looking for is to limit drawdown while keeping a reasonable winning percentage and profitability.
This is where most commercial EAs fail. They are over-optimized on old data, and the market doesn't repeat itself. If the optimization data used is very recent, they may work well for a little while. So far, all those "never lost a trade in 10 years" EAs end up mysteriously losing multiple trades as soon as they are sold.
The next thing to do is look carefully at your data from your backtests. Is there a certain time of day that the EA fails more often? Perhaps the strategy works best only during the London trading session. Maybe it works well except during the New York session. Find out. If you do see a consistent pattern like this, program the EA to not open trades during those times.
Does the EA consistently have issues on certain days of the week. Maybe it wins more on Fridays and loses more on Tuesdays. If so, lock out trading on the losing days.
Look at how it performs around major news releases. If it gets nailed by Non Farm Payrolls, tell it to stop trading for a few hours around that release. You could even put in a user selectable feature where the major news events for the upcoming week or month are entered to lock down trading at those times.
Build in a volatility indicator or two. Then test settings of that to determine minimum and maximum volatility for the EA to perform well.
Your EA will now trade a lot less, but will trade at the best days and the best times for the strategy you devised. It will also skip trading around news events that could upset the strategy and will avoid trading when price action is too slow or too fast.
Some EAs do not set a SL, and instead have a "hidden SL" to keep the broker from trying to hunt the stops. A hidden SL is a good thing, but you really need a SL to be set. This protects you if the EA gets turned off, the computer gets turned off, or the connection to the MT4 server is lost for any reason. If you do use a hidden SL, set a hard stop farther out. Put it far enough away to make it hard for the broker to spike you out, but close enough so that the account won't be seriously damaged if all open trades hit the SL at once.
Set up a dozen or so demo accounts and let it run with different settings. If it's still doing well after a month or two, you can think about opening a small live account. using the best settings.
