Has anybody developed an EA like what I did? | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Anybody know a site like forexfactory but for equities? 0 replies

Pending Order EA (Does anybody know one like this?) 1 reply

Has anybody had any experience with Delta Stock? 61 replies

Has anybody ever lost a scalp... 12 replies

Has anybody trade the german IFO? 7 replies

Attachments: Has anybody developed an EA like what I did?
Exit Attachments

Has anybody developed an EA like what I did?

  • #1
  • First Post: Sep 10, 2016 8:01pm Sep 10, 2016 8:01pm
Hi Guys,
I know there are lots of professional traders here, where is the good place for my confusion.
I've developed an EA based on broker's API. This EA contains more than 200 EA entities, every one of which works on one instrument's one timeframe with one technical strategy. With this method, my EA now can trade on about 50 instruments (including almost all the currencies, CFDs and indexes) on 9 timeframes (m5, m15, m30, H1, H2, H3 ,H4, H6, H8).
Let me briefly describe how I generate one EA entity.
1. Select the technical strategy, I named it EA1 (for example, the strategy is in timeframe X, if MACD gold cross and in timeframe Y, MACD DIFF goes up, then generate buy signal).
2. Run selection program of my own to determine which is the best timeframe of X and Y and also select the best StopLoss level.
Eventually, let say for GBP/USD, we found that with EA1, if X is H4, then the best value for Y is H3, the best maximum stoploss is 250 pips. Then I create an EA entiry for GBP/USD working at timeframe H4 with those paramters I selected.
With these method, I totally select more than 200 entities, which can work simultaneously, generating 14 signals per day on average. Of course, with a very good backtest return from year 2002~2016.
BTW, with every signal, use 1% of the equity to open the position (leverage 1:100), and maximum used margin less than 50% of equity.

So is this the correct way do select EA parameters? I am doing it with curve fitting?
Inner Trader
  • #2
  • Sep 11, 2016 8:58am Sep 11, 2016 8:58am
Try the monte carlo simulation on your system to see how robust and stable it could get...Curve fitting is a very dangerous thing to do...I tried doing it before, sooner or later leads to margin calls on live accounts.
Simplicity is the key to brilliance.
  • #3
  • Sep 11, 2016 11:04am Sep 11, 2016 11:04am
Quoting Inceptionist
Try the monte carlo simulation on your system to see how robust and stable it could get...Curve fitting is a very dangerous thing to do...I tried doing it before, sooner or later leads to margin calls on live accounts.
Yes, Curve Fitting is my most concern. I've run this EA on some live accounts for about 2 months, although no margin call, its return is either not as good as that in backtesting.
Never try monte carlo simulation before, is it really useful for us? I think its input is random, which can ruin any of our trading system with specific rules.
Inner Trader
  • #4
  • Sep 11, 2016 2:11pm Sep 11, 2016 2:11pm
I am working on something similar, an algorithm which uses 400+ indicators that can work on multiple timeframes and multiple assets. But putting a lot of emphasis on support/resistance as well. Should be finished with it soon.
  • #5
  • Sep 11, 2016 2:37pm Sep 11, 2016 2:37pm
Quoting PoundPotamus
I am working on something similar, an algorithm which uses 400+ indicators that can work on multiple timeframes and multiple assets. But putting a lot of emphasis on support/resistance as well. Should be finished with it soon.
Good to know that. Maybe we can discuss something based on this, like how to reduce the MDD with multiple timeframes/multiple assets, how to determine the money management, etc.
Inner Trader
  • #6
  • Sep 11, 2016 4:49pm Sep 11, 2016 4:49pm
Quoting VEEFX
{quote} Interesting James. This is my topic of 'research and development' for past two years. I now have a single EA running on 21 currency pairs and across all timeframes (MN thru M1) using four MT4 instances running in parallel on a high powered laptop. Similar to yours, my entries are consistent across all pairs/timeframes but using pure PA with no technical indicators or setups. My edge is in the exits (i.e profit and loss management) so let's leave that out of the discussion. I have been forward testing for over a year with gathering stats...
Thanks Vee, for question 1 and 2, it is because I am developing my EA actually from scratch based on brokers' API. I can do whatever I want, I programmed my own backtest program which can support any number of EA's backtest for the same account.
Lots of brokers have their own API providing historical data interface and trading interface, like FXCM's forexconnect API, Dukuscopy's JForex, etc. We can use C++/Java do the developement, not the language from MT4/MT5.
Because I am doing backtest on only one account, its MM is much simpler. For example, for a $10000 equity account, for every signal, 1% of the equity is used for margin, that is 100, roughly about 10,000 (0.1 lot)position is opened with leverage 1:100. With this, totally maximun, $5000 can be used as margin (if the equity is fixed as $10000).
Inner Trader
  • #7
  • Sep 11, 2016 5:55pm Sep 11, 2016 5:55pm
Quoting VEEFX
{quote} Thanks. When the time is right, I will need to hire someone of your skills but my API solution will have to be broker independent... something similar to FIX. Not there yet! Your MM is similar to mine but my model requires more leverage and I find 2x (2% risk) works better for my setups. I also vary my risk, pair and timeframe selection based on liquidity and volatility. Under 50 open positions slows down my yield curve. Do you hedge? Do you add to positions (hopefully to the ones in profit)? I see you are from USA like me but trading from...
Hi Vee, I think it is not hard to reprogram your EA with broker's API based on your current MT4 EA. Maybe 40% of the code can be reused. I finished my first version from scratch within 1.5 months. I think if we want to run EA like what we do, API is the best choice, MT4/5 is not applicable.
Although there is no obvious logic in my EA handling hedging and position adding, they do exist when all the EA entities are running. For example, a long position of EUR/USD was opened with EA1 (entering at the tf of H8) yesterday, but a short position of EUR/USD may be opened with EA5 (entering at the tf of m30) today, also another long position with EA19 (entering at the tf of H4) may also be opened today. This includes both hedging and adding.
My EA runs 7x24, no manual intervention except that I want to update the EA program itself.
Inner Trader
  • #8
  • Sep 11, 2016 6:09pm Sep 11, 2016 6:09pm
Quoting VEEFX
Looking at the stats, you take frequent loss to improve your RR. I used to do that but when I altered the entire perspective on frequency of profit taking, I had a big revelation. Just curious if you tried various profit taking schemes instead of the traditional SL and TP per trade? Try it and you will be amazed (Hint: I do not mean take smaller profits and larger losses to reverse your RR) PS: I delete my posts in 24 hours to avoid clutter.
Yes, my trade profit/loss percentage is about 38%, so lots of loss trades. My profit comes from "cut loss, hold profit as long as possible", you know, from backtesting result, the average holding days of loss trade is about 1.7 days, the average holding days of profit trade is about 6 days.
As you know, I only set SL, never set TP, the only exit condition is the technical singal and SL.
I did try to increase the win percentage, but no luck, the total profit didn't increase accordingly.
One of my concern is that for those 254 EA entities, I use different SL setups and different exit timeframes, which are all selected by my own selection program running on past 14 years data. Still don't know if there is too much curve fitting or not.

Thanks for your suggestion, Vee, I think I can try a little more on the TP strategy.
Inner Trader
  • #9
  • Sep 11, 2016 7:50pm Sep 11, 2016 7:50pm
Quoting VEEFX
{quote} It is curve fitting if you are using stats from history to anticipate future price movements or manage risk. This is where we start to differ BUT we better stop talking as there is a HUGE edge about to be disclosed with this approach.
Yes, curve fitting is a big topic, which is what I try to prevent from the beginning of EA design.
I only use historic data to generate two parameters of my EA, one it the maximum SL, the other is the exit timeframe. I treat the SL range and suitable timeframe as somewhat constant character of an asset. Hope I am taking the right way.

Yes, all my trades are being handled by my EA. Actually, I haven't opened/closed any postion for two months.

Although I've used the EA for my live accounts for more than two months, I still don't have too much confidence. This is also why I posted here hoping somebody may find out my big risk before I blow up my accounts.
Inner Trader
  • #10
  • Sep 11, 2016 9:26pm Sep 11, 2016 9:26pm
Quoting VEEFX
{quote} Two months? Your TE live account is only 4 days old so hard to assess if you are going to blow up. My speculation is you will at some point very soon as you have not gone thru enough forward testing in demo (assuming) and relying on your backtesting is a faulty approach imo. I have been running my model for almost past 14 months now with constant tweaks and features and building of tools that exposes the 'hidden' nature of currency markets. I will say this with a high level of certainty. Get rid of indexes and non-currency pairs from your...
Oh.. I have some other accounts running my EA for more than 2 months.
You are right. I need to be more focusing. It is a long way for me, I just started.
Yes, every EA has different technical setup, mainly different SL range and different enter/exit timeframes.
Vee, what do you mean about the consitency? Am I missing something?
Inner Trader
  • #11
  • Sep 13, 2016 6:52pm Sep 13, 2016 6:52pm
Quoting VEEFX
{quote} James - I am not surprised with the lack of responses in this thread. When I embarked on my journey of automating market behavior tendencies, I tried to find help and I could not find any strategies or anyone who came close to what I wanted to accomplish. It has taken me a long time to learn mql and forward test in baby steps due to lack of multipair back testing tools available. I doubt you will find much participation here as this type of approach is suited for hedge funds and not retail traders. Curious from the back testing if you have...
Vee,
You are totally right. Till now, you are the first one understanding what I am doing in the past 4 months. Thank you.
Actually, my strategy is pretty simple, like what I've been using for the past several years. It's something like MACD golden cross/dead cross, just with some guarding conditions. It is a trend following strategy. Here I am happy to share you some of my findings from my backtesting.
1. The best instruments for my trend following strategy is those related to GBP and JPY, like GBP/JPY, EUR/JPY, GBP/USD.
2. With just one pair of instrument, the backtest result of 14 years(2002~2016) is just so so. Most of my EA entities only have 80%~150% return over 14 years. Only few of them with more than 10 times. (3% margin per trade)
3. For single pair backtesting, if I increase the used margin percentage per trade, like 20% percentage per trade, Almost all of the EA entities will blow up the account somewhere from year 2002 ~ 2016. But the most interesting thing is that, if I running more than 200 EA entities together, with average used margin percentage about 40% at all times, it never blows up the account, and the maximum equity drawdown is about 60%. I think that's what the word "hedging" means.
Inner Trader
  • #12
  • Sep 13, 2016 8:05pm Sep 13, 2016 8:05pm
Quoting VEEFX
{quote} Thanks for the details James. Where we completely differ is in the use of technical indicators like MACD. Have you tried pure PA for your entries? Is it possible to share a same result of your backtesting (anyone or anything is fine)? just want to see if it matches closely to fxbook or TE format in regards to the performance characteristics. Perhaps there is some room for us to collaborate without exposing our years of research. I have one custom indicator (a single function in MQL) which will return one of the four states of up and down...
My EA is a standalone program running on Windows/Linux. I think it can not call MQL based indicators/exe, which can only be called from MT4.
Of course I can share one of my backtesting result.
Attached File(s)
File Type: xml ea_GBP_BT.xml   2.0 MB | 373 downloads
. But it is in my own format. I just generate the backtesting result and import it to Excel to do analysis.
I know PA is a good thing, but I've already stopped learning new technical thing some years ago. So I know nothing about PA. :-).
Thanks Vee.
Inner Trader
  • #13
  • Last Post: Sep 13, 2016 8:32pm Sep 13, 2016 8:32pm
Quoting VEEFX
{quote} Thanks. Back to paving our own path then. Good Luck.
Gook luck, my friend. Nice meeting you.
Inner Trader
0 traders viewing now