Forum Replies Created
-
Posts
-
Popov says it’s normal for an added commission to degrade the strategies performance to a level that makes it twice as difficult to pass the acceptance criteria. I’m going to assume that’s the biggest reason why I’m struggling with custom symbol settings.
As a test tonight, I ran two browsers at the same time using MetaTrader-Demo, EURUSD, H1–one with custom symbol settings, the other with no changes. The only changes I made under symbol settings was swap long, swap short, and a $3 commission. After running both for about 45 minutes, here were the results…
Original:
Generator
Generated strategies 104006
Passed validation 51 (0.05 %)
Normalization
Calculated strategies 51 (3769)
Normalized strategies 49 (96.08 %)
Passed validation 51 (100.00 %)
Monte Carlo validation
Calculated strategies 51 (1020)
Passed validation 17 (33.33 %)Custom symbol settings:
Generator
Generated strategies 104396
Passed validation 28 (0.03 %)
Normalization
Calculated strategies 28 (2105)
Normalized strategies 28 (100.00 %)
Passed validation 28 (100.00 %)
Monte Carlo validation
Calculated strategies 28 (560)
Passed validation 9 (32.14 %)They run about the same speed, but one has half the results. It seems to be even worse if I upload data from another broker (since MetaTrader-Demo only has a handful of currency pairs). Is this maybe because of the $3 commission, where the other doesn’t have one at all?
Oh, another observation, the EAs did NOT close out Friday trades properly at the time they were set. I double checked the EAs to be certain:
// Session time is set in seconds from 00:00
const int sessionSundayOpen = 0; // 00:00
const int sessionSundayClose = 86400; // 24:00
const int sessionMondayThursdayOpen = 300; // 00:05
const int sessionMondayThursdayClose = 86400; // 24:00
const int sessionFridayOpen = 300; // 00:05
const int sessionFridayClose = 86100; // 23:55
const bool sessionIgnoreSunday = true;
const bool sessionCloseAtSessionClose = false;
const bool sessionCloseAtFridayClose = true;The broker closed at 23:59 and trades are still open.
Hey Petko!
FTMO is very legit, one of the most respected prop firms. Hence why their challenge is notoriously difficult, they are around for the long term and won’t fund a trader with bad risk management, haha! They aren’t a broker persay, they use LMAX.
I took all 28 major currency pairs and split them into diversified groups, and then started generating strategies on each group of six at a time, since I couldn’t do more than that at once with my CPU without slowing down the reactor. I’ve been trading on most of the 28–some I couldn’t get enough strategies for. So I did trade the ones you mentioned.
The spread, swap, and commission with LMAX look better than most other brokers I have used in the past. EURUSD for example currently has a spread of 2, commission is $3, swap long is -0.91 and swap short is +0.24. I’ve entered these figures into the symbol settings for each pair that I’m generating strategies for. I don’t know why these symbol settings would be bad, but maybe this is a key to why I’m getting so few strategies, I dunno…
On the demo account, FTMO is passing along data from LMAX, who is only providing 2 weeks of data for demo accounts. There is nothing they or I can do about it from the answer I received. Maybe I could purchase it directly from LMAX, but they don’t seem to want to deal with individuals. I tried getting more historical data from LMAX via MT5 because it is listed as a demo server on a native install (not from any broker, just direct from metatrader5.com), but the data they provide there is nowhere near the same or as good as what they give as a liquidity provider to large firms like FTMO (spread, commission, etc.). So I’m still stuck downloading data that’s not identical and trying to match it up via symbol settings. If I had an account with them long enough, I could use my own captured data over a longer time period, but I need to pass the challenge to get a real account to have that, LOL! 14-day demo accounts won’t give you more data than what I’ve got.
One problem I’ve had is opening too many trades at too high a lot (0.05 or higher is TOO much!) and easily going below 5% equity. So I’ve had to stick with 0.01 lots because the stop loss on most of these strategies is quite large, and if you have 50 trades open and 2/3 of them start going in the negative, you can very easily hit your daily maximum loss. Is there a way to limit how many trades each EA opens at once? Maybe having only 5 strategies per pair and limiting myself to 10 pairs would help limit max open trades at once and risk, allowing a larger lot size. If you get 10 strategies per pair, you are back to 0.01 lot size to avoid max daily limits. The hedging really cancels out a lot of winning trades. It’s really tough to hit +10% profit in a month without ever risking more than 5% total equity in open trades at once I’ve discovered. I started using a 3rd party EA to enable a better trailing stop on all opened trades to capture positive equity and close it out faster, since I wasn’t hitting the take profit very often without dropping into negative equity a lot. I’ve tried changing the reactor to accomplish this, either with a trailing stop, or with a stop loss smaller than take profit, but wound up with even fewer strategies.
I think I need to start all over generating strategies on another demo account with another broker just to be sure symbol settings is what is preventing me from finding good strategies. I’ve wasted about 2 weeks of CPU time so far. The strategies I’ve been generating are losing more than winning, even when they look good on the last 2 weeks of LMAX data before putting them on the demo account for forward testing.
I have not been able to post with attached photos, they get flagged for moderation, but the moderators can’t find the post afterward to approve it. So I’ve had to skip attaching photos if I want the posts to appear.
Thanks John! You should try it too! I did try the 2 week demo and failed, the draw down was too much. For example, I’m running 5 portfolio EAs on 5 currencies, my equity is positive $369, but my margin is almost $1200 because so many trades are open with hedging. So that’s almost half way to my max daily loss of $2500 on the 2 week demo. Worse, I had a positive equity over $1000 yesterday, but it dropped to $300 something (still positive) and yet they consider that a $700 negative toward your max daily loss. So I’ve had to run another EA that gives a trailing stop on all open trades to avoid this in the future. At any rate, I failed my first attempt on a free 2 week demo. I also did not have very good EAs as I was only using about 100 trades to generate them faster. 😉
So, here are my current settings. Happy to hear any advice on them!
I’m using MetaTrader-Demo as the data source, but going into Symbol Settings and making it match my broker. I’m going for 15M timeframe currently on the 28 major pairs. My acceptance criteria is:
- Minimum count of trades: 300 – 500 (I’m still experimenting)
- Minimum balance stability: 80 – 85
- Minimum profit factor: 1.2
- I’ve also tried Minimum r – squared of various numbers from 60 – 85 instead of balance stability. Any suggestions?
For both in sample and out of sample I’m using Minimum profit factor: 1.1
For Monte Carlo Validation I’m using the default of Minimum net profit: 10, and Minimum count of trades: 100. I might try Minimum profit factor: 1.1 instead, would love suggestions!
Under Reactor, my Strategy properties are:
- Entry lots: 0.1
- Stop loss: always use
- Type: Fixed
- Min pips: 10
- Max pips: 100
- Take profit: Always use
- Min pips profit: 5
- Max pips profit: 50 (I’ve tried 10 and 100, but I was hoping trades would close sooner this way, since they must close by Friday anyway)
Generator settings:
- Working minutes: 0 (I run until I get 10 or so good looking strategies for a portfolio–IF it finds any at all. I’ve let it go to 100 strategies with fewer minimum trades, but they were failures in demo accounts)
- Search best: Net balance
- Out of sample: 20% OOS (I’ve also tried 30%, not sure which is better)
- Max entry indicators: 4
- Max exit indicators: 2
- Generate strategies with preset indicators is NOT checked
- Use common acceptance criteria IS checked
I’m not using anything for optimize strategies. They took too long and rarely passed with less than 500 trades, and even then didn’t usually pass Monte Carlo afterward, but would pass Monte Carlo if they were not optimized. So I decided not to use optimization at all, unless someone has better insight?
The only thing checked under robustness testing is Monte Carlo, with the previously mentioned settings. Multi Market seemed to fail many strategies that otherwise looked good, and I’m barely getting 1 – 3 strategies in 48+hrs as it is. Sigh…
Any suggestions to any of the above?
BTW, here is what one of the analysts on the FTMO team had for an experience with algorithmic trading, taking four months to generate strategies and then a year of trading them. They may have made several flaws with their methodology, including not constantly generating new strategies and swapping them out, but this is why I want to prove that it can be done successfully and long-term with EA Studio! 😉