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

Matthew Roberts wrote:

update: it might be possible to load EAs via templates...just found this link on Quora:

I was just messing around with this approach and it does seem to work - I had it launching a new chart window at a specific time with a custom EA added to it, with my launch time parameter correctly modified.

The key is to pre-prepare your templates and have them ready to go, and your script would programatically launch a new daily template on a specific time, and that template would reference the latest EA (which would have been added a few minutes ago)...if that makes sense.  One way would be to add a date parameter to the filename - such as _dd.ex5 and then have a template per day.

Obviously that doesn't get around the manual part for loading the collection into EAS and exporting a portfolio.

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

Any thoughts on generating strategies with a higher risk to reward ratio? I'd like my winners to be bigger than my losers on average, even if it's only by 1.2x. I've tried using a larger TP than SL when creating strategies, but when running live the indicator's exit strategies close trades before I can achieve it, often one quarter or less than my stop losses, which requires an unachievable win rate to be profitable long term. I do think there is merit in "over optimizing" for a shorter time period and re-generating new EAs daily. I had a couple good days this week, but the build up of negative equity long term is killing me on prop firms that require closing all trades for the weekend, so I'm trying to improve the risk to reward ratio.


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

Hello, I've only been using the commercial algorithm for two months and I was the same as timelleston. I had spent weeks considering how to create a portfolio, learning about EA Studio and creating systems. I did not understand how I had systems in various temporalities and pairs that worked on IS and OOS and the second day of testing them in demo they gave catastrophic results. I have read the topic from top to bottom, in some parts I have read it two or three times. (Also the google translation is not the best) and I just want to say thank you. Thanks to all of you who have contributed your grain of sand but above all thanks to timelleston and sleytus. To sleytus for the knowledge taught but above all for teaching us that more data is not better and to timelleston for his interest, proofs shown and making the thread move forward. You've all done more than that but I'm not going to get romantic.

I have taken your conclusions with which I agree and I think I have found something, and if I am not mistaken something very good and how I have come this far thanks to you I would like to share it, although I have not done more than 20 tests I think there is a clear advantage. Although it is best that you try it since I am not going to show any results, since I have been with the portfolio in demo for a day, but in backtests with similar strategies and "end date" in the past to see results in the future they work a time.

As sleytus says, the strategies (and especially the ones we were testing, I have also tried everything shared in the topic and a hundred more combinations). As he was saying, the strategies are very adjusted to the sample and the market is changing. And these two go hand in hand. I have only made the market less volatile by raising the time frame to 1H. Why? The candles are more summarized and this means that we collect more data in the same number of bars, we minimize the noise that we talk about so much. For example, if we take 500 candles in 15 min and we are in a strong uptrend, it is very likely that the trend will change its volume and the system will start to fail. If we do it in 1h, can the trend catch us and what happened before it and we eliminate x4 noise?

And I have worked with data of 6175 bars, which has been a year in 1h. With less it has not gone well for me and I have been going up until I found that one. It is obvious that it can be better adjusted to this way of operating. I'm going to stop spinning and I'm going to say the settings.

I work with 6175 bars from my broker.
EUR USD in 1H (one year)
short and long
reverse signal
Fixed or trailing stop of 200
No take profit

Better return/drawdown
in the sample
Max indicators 2 and 2

Walk Fordwalk:
Segments: 4 Validated segments: 3
30% oos and 20 steps

Monte Carlo:
Tests: 20 and 50% validated.

Criteria of acceptance:
Minimum net profit: 100
Min. number of operations: 200
Max Drawdown: 12%
Max Stagnant days: 60

Then I leave a day and something else for the reactor to think about and order the collection by SQN. This puts us at the top of the operations with less DD and they rise exponentially, that is, if the sample is one year, in the last six months it has been working better and I remove those that currently (an imaginary 10% OOS) are in DD . Because I only want systems that are currently winning, that my system has worked in the last year but in the last six months better. And I normalize, (this above all touches our stop of 200 in many cases it is very big.)

In the tests I've done, none of them crash like we used to, the systems that don't work stay in range for a bit and then lose and the winners win. From there I made a portfolio of at least 20 experts for my opinion and remove the stagnant or losers and leave winners. Considering that our systems have a great curve for the short amount of time. But in this systems management I am very new, as in all this. I am 26 years old and I have never made money consistently but I think I have improved somewhat the results we had. I would like people to try it and modify this concept to find a more optimal way if possible.