Forex Software — Definition of insanity...looking for some guidance

Re: Definition of insanity...looking for some guidance

2024-12-28T17:18:13Z

The thread is a bit older, but maybe someone is still interested.
I think almost everyone here has had these experiences.
After spending a lot of time with FSB and experimenting extensively, I have a strong suspicion that it is a systemic issue with FSB, which essentially occurs inevitably.

I suspect this is partly related to how FSB generates strategies. Essentially, it brute-forces indicator values, which are heavily dependent on backtest data and cannot respond to future events.
It always leads to curve fitting.

Ideally, FSB would need to combine strategy generation and trade execution while automatically identifying patterns to determine which strategies or parameters actually work live and which do not. In my opinion, this is not feasible manually or by human effort.

Additionally, the indicators supported by FSB are quite limited. Many indicators are simply not good enough – even if you adjust the parameters, you won’t get a robust strategy or algorithm.

Therefore, at least for me, FSB (as it currently is) is not of interest.

Re: Definition of insanity...looking for some guidance

2024-09-29T11:18:05Z

I recovered this post and made some tests based on the info provided in page 2 (reported below), on CADJPY M15, with 3 different In Sample Period among August and September 2024.

The only differences were in using EA Studio OOS analysis by changing the data period, and by filtering strategies separately also for Win/loss, not only for SQN.

The OOS period was immediately following the In Sample Period: for example, if In Sample ended on September 1, OOS period started on September 1.

Unfortunately, in 3 different periods, in 5 of 6 scenario (filtering only with SQN and then only with WL) lose money in the first day, except one case.

I attach image for better understanding.

timelleston wrote:

Ok, so my End Date Limit Data Horizons are as follows (each with 300 bars):
7 Sept
10 Sept
15 Sept
21 Sept
26 Sept
That gives each run 300 bars back in time from those dates, which almost overlaps on the start dates.
Strategy Properties:
Entry Lots: 0.1
Direction: Long and Short
Opposite Signal: Ignore or Reverse
Stop Loss: Always use, Fixed, 95 & 95
Take Profit: Always use, Fixed, 95 & 95 (I'd normally change these a bit, but these will do for the test)
Generator Properties:
Search best: Net Balance
Out of Sample: In sample
Max Entry: 4
Max Exit: 2
Validation: Use common acceptance is ticked
Common Acceptance is:
Complete Backtest:
Min Win/Loss: 0.6
Min Net Profit: 10
Min R-Squared: 70
Cooking 5 portfolios now for 120 minutes...
Once cooked, I'll set Sort Collection by System Quality Number, and move the Top 50 to Portfolio, then download into MT5 and run backtest on Every Tick based on real Ticks, using the same start dates as the Portfolio was generated on, but the end date will be 29th Sept...so we get the "real-data" into the portfolios.
Cooking now

Re: Definition of insanity...looking for some guidance

2024-03-10T14:48:51Z

Good afternoon everyone,
found this interesting topic in testing on just 300 bars.

Before reading it all, I started generated with EA Studio for EURAUD based on the The5ers prop account data.

Had no success after a couple of portfolio, for where I need to close trades manually and got huge drawdown the first day already.

I then saw some posts talking about CADJPY and GBPJPY, @sleytus and @timelleston are still they your best pairs now?

At march 2024, how much bars are you using for generating?

Would love to know your updates and overall result.

My current approach for data used and strategies selection (to put them in the portofio):
500 Bars
No OOS
Small optimization only on SL/TP (+-10)
Not always: Preset indicators: 3 bars rising (BB) + MACD live above signal line (both for trade with the trend)
M15
WinLoss ratio 1.0
Minimum trades 10
SL 10-30 pips
TP 5-30 pips
Drawdown: max 10usd (testing 0.1Lts with 5k)

Do you keep 2 correlated strategies in the portfolio or do you remove them? And how (trend, similar rules, both)?

Please share any feedbacks and updates you feel!
Thanks

Re: Definition of insanity...looking for some guidance

2023-05-08T21:27:09Z

Thank you very much for everything, I'm glad to discuss this with interesting people and share. I closed that account to take the systems to real and they have more or less the same results. It's also worth admitting that in January I entered into a trade on the DAX that yielded a very good profit. I've noticed that indices provide better and more consistent results.

My portfolio usually has around 8-12 advisors. Since March, I have eliminated 6 of them and replaced them with others. Let's say my "trick" is to have advisors who are winning or not in a drawdown every month. That's why I'm confident that in the future, I will use more historical data so that I don't have to be so vigilant and avoid trading based solely on system performance.

I completely agree with you about the importance of data. In truth, from this post, I have learned a lot about data, overfitting, and other related concepts. However, I got confused in the beginning when I read here that to avoid overfitting, one should use 200 bars. I don't think anyone achieved consistent results with such a small dataset, right? I believe a system with battle scars is more stable than one without. When I saw your message the other day and noticed that you were using data since '87, I became curious. Although I don't have that much data, I started experimenting with more extensive datasets than I usually use, and I was surprised by the results. I think I'll be testing it in the coming weeks. Additionally, I was surprised, or perhaps I didn't understand correctly, that you have the generator running constantly. Does that mean you have a VPS with EA Studio?

Thank you, and I apologize if something is not understood well. I am translating through ChatGPT.

Re: Definition of insanity...looking for some guidance

2023-05-08T01:00:17Z

Congratulations on your remarkable results! I am curious to know how your portfolio performed during the months of March, April, and May (up until now).

I'm delighted to hear that your method is yielding positive outcomes for you as well. My approach, which I find to be rather stress-free, involves running the generator continuously, reviewing its findings approximately every two weeks, subjecting it to rigorous stress tests, conducting six years of out-of-sample testing, and only implementing it live once it successfully passes all evaluations. It's a straightforward process.

Based on my research, I have observed that the longer the data horizon utilized, the less susceptible the systems are to curve fitting. This is primarily due to the difficulty of fitting such vast datasets through random chance alone, thereby enabling the models to generalize across various market conditions. Consequently, this enhances their chances of survival in the future.

I'm eagerly looking forward to witnessing your progress as well. Wishing you the best of luck!

Re: Definition of insanity...looking for some guidance

2023-05-06T14:35:49Z

First of all, I would like to congratulate you on your admirable results, great work both by yourself and EA Studio. I would bet that, with your method, you will achieve good results in the medium and long term. My systems and the way I manage them are completely different. I use a much smaller sample (7500 4-hour bars) and one year of out-of-sample testing. I BELIEVE this yields systems with better or the same results in less time, but at the same time, they are less robust, and for that, you need to have a well-diversified portfolio and remove those that decrease their results and replace them. As I mentioned, in my opinion, my way of trading is stressful because one of my systems can last much less than one of yours, or at least I believe so. Undoubtedly, if I had the capital, I would use your method because I would live a more peaceful life. As with everything in life, there are a thousand ways to do the same thing, and sharing is how we grow.

I want to emphasize that I am very surprised by your results. I wouldn't have believed that using such a large sample would yield those results. As always, our beliefs limit us.

the track record that it has achieved in the past few months following this method:
https://ibb.co/NxDHXTx

Re: Definition of insanity...looking for some guidance

2023-05-02T16:03:11Z

As a long term quantitative trader (doing this since 2008), I follow a rigorous methodology to ensure that my trading strategies are based on robust data analysis. Specifically, I utilize a vast dataset spanning from 1987 to 2017, with a limiting end date as yourself. Subsequently, I retest the resulting strategies on a more recent dataset from 2017 to present. Upon passing stringent stability tests and meeting my performance metrics, I promptly deploy the model into the live trading environment. I have found that this approach is most effective for obtaining consistently favorable outcomes. Conversely, I have found that attempting to use short timeframes such as days, weeks, months, or even ten years of data does not yield any satisfactory results, it´s just almost always randomness with short datasets and you simply can´t tell if there is a real edge with such short horizons. Only by incorporating my comprehensive Forex dataset spanning back to 1987 am I able to achieve the stability and profitability.

Re: Definition of insanity...looking for some guidance

2023-05-01T16:18:03Z

The comment I posted above, I only used it for a few weeks. What I currently use to determine if a system is currently functioning is to generate it "In sample" with an end date of X in the Data section. In other words, I create the system with past data, once I have enough EAs loaded, I remove the end date and click Recalculate. Now I truly know how it behaves in reality without the IS and OOS of EA Studio.

I don't use as few days as what is being looked for here, although I would love to have a combination that gives good results. But that's my way of eliminating 95% of systems that don't fit well in the new market.

That's how I win, if someone wants to share something even in private.

Re: Definition of insanity...looking for some guidance

2023-02-10T19:40:32Z

My bad, no exporting function in trial mode.

Yes, the max open position feature is an input for portfolio EAs.

Re: Definition of insanity...looking for some guidance

2023-02-10T18:40:09Z

footon wrote:

Yes.
You're not trialing? In trial version you should have full functionality also, gives you an idea how it works.
GPyles wrote:
geektrader wrote:
This doesn't exist for single EA Studio EAs, they don't have this option and that's what he was referring to.
So if I purchase a license then I will have this option with the Portfolio EA's, correct?

Yes I am using the trail version. In the trail I cannot export Portfolio EA's. I am assuming that the option only exists in the Portfolio EA and not in the individual Ea's...

Re: Definition of insanity...looking for some guidance

2023-02-10T17:26:38Z

Yes.
You're not trialing? In trial version you should have full functionality also, gives you an idea how it works.

GPyles wrote:

geektrader wrote:
This doesn't exist for single EA Studio EAs, they don't have this option and that's what he was referring to.
So if I purchase a license then I will have this option with the Portfolio EA's, correct?

Re: Definition of insanity...looking for some guidance

2023-02-10T15:59:33Z

geektrader wrote:

This doesn't exist for single EA Studio EAs, they don't have this option and that's what he was referring to.

So if I purchase a license then I will have this option with the Portfolio EA's, correct?

Re: Definition of insanity...looking for some guidance

2023-02-10T11:39:04Z

This doesn't exist for single EA Studio EAs, they don't have this option and that's what he was referring to.

Re: Definition of insanity...looking for some guidance

2023-02-10T11:16:29Z

In the expert inputs while you are attaching it on chart.

GPyles wrote:

sleytus wrote:
Very, very cool -- thanks for performing the tests...
If you compare different pairs I think you'll find that CJ lends itself best. GJ also works, but sometimes DD can get a bit hairy...
EA Studio is the magic sauce since it allows us to create new portfolio EAs at the press of a button. Even if a portfolio EA could survive a few days I still substitute with a new one every day starting a few hours prior to London Open. And if there are Open trades then I set Max Open Trades to '0' for the decaying EAs and let those close gracefully. I also use broker data because it is more up-to-date. Premium Data lags behind by about 4 hours -- and because my data horizon is so short then every hour counts...
One last point -- I only provided very crude instructions -- i.e. CJ / M15 / 300 bars. There are many other configuration settings in EA Studio where we probably differ. Yet, you still saw decent results. That is an indication this approach is robust and doesn't necessarily rely on fine tuning...

How do you "set Max Open Trades to '0'"?

Re: Definition of insanity...looking for some guidance

2023-02-10T00:10:50Z

sleytus wrote:

Very, very cool -- thanks for performing the tests...
If you compare different pairs I think you'll find that CJ lends itself best. GJ also works, but sometimes DD can get a bit hairy...
EA Studio is the magic sauce since it allows us to create new portfolio EAs at the press of a button. Even if a portfolio EA could survive a few days I still substitute with a new one every day starting a few hours prior to London Open. And if there are Open trades then I set Max Open Trades to '0' for the decaying EAs and let those close gracefully. I also use broker data because it is more up-to-date. Premium Data lags behind by about 4 hours -- and because my data horizon is so short then every hour counts...
One last point -- I only provided very crude instructions -- i.e. CJ / M15 / 300 bars. There are many other configuration settings in EA Studio where we probably differ. Yet, you still saw decent results. That is an indication this approach is robust and doesn't necessarily rely on fine tuning...

How do you "set Max Open Trades to '0'"?