Forex Factory (https://www.forexfactory.com/forum.php)
-   Trading Systems (https://www.forexfactory.com/forumdisplay.php?f=71)
-   -   My "Ultron" EA for GBPUSD H1 timeframe (https://www.forexfactory.com/showthread.php?t=840339)

robots4me Nov 17, 2018 11:44pm | Post# 361

Do not punish us with all this posts having no contribution to this thread at all
I agree. And if you review the sequence of posts then you'll see how things transpired and began down a different path...

Before @CashBoa announces his 4000% return the thread is progressing quietly. He then announces his 4000% return EA. Even though MT4 automatically generates the corresponding stats for each chart, for some reason @CashBoa chooses not to include those. Why? Who knows.

Then people start to question 4000% return and also inquire about more information and provide feedback for additional features. Now, the thread has become focused on @CashBoa's new 4000% EA and he thinks this is all really "exciting" (his words). It doesn't occur to him that he is wasting people's time. He is also contributing to this misguided concept of a Holy Grail.

When I earlier said that forex was "difficult" -- I didn't mean physically difficult. What I meant is it is difficult to succeed -- and I don't think anyone would dispute that. Most all statistics and surveys suggest the overwhelming majority of retail traders do not make a profit. I think two major sources of so much failing is: (1) misinformation, (2) expecting a holy grail strategy to save you from paying attention to details.

Though @Pawel may have written the EA in a few hours, he clearly is someone who pays attention to details and how the market works -- especially the GBPUSD pair. And he applied that experience and translated an idea into an EA. @CashBoa has neither -- but, because he announced a shiny new object then, of course, a number of people will be attracted. In the end -- it wastes everyone's time and we end up writing posts that have nothing to do with Ultron or learning more about trading.

One last point -- as far as my posts discouraging @CashBoa, just the opposite is true. Keep in mind he made his announcement without any additional details -- so, clearly, he is not shy. If anything, my skepticism would motivate him to prove me wrong and himself, right. So -- don't worry that he might actually develop a 4000% return EA and not let you know about it. If he were to succeed, you will definitely hear about it.

I think I've exhausted my allotted time -- I'll try not to say anything more about 4000% return EAs...

3ndlessHope Nov 18, 2018 2:46am | Post# 362

{quote} I was taught to focus on one thing. Each cross is different and each market is different. You cannot be good at everything. You basically told two things that are opposite to each other. If markets are all the same, why EA works great only on one? Also, I have stated my opinion about so-called "diversification". If you diversify your trades, it means that you are not sure what you are doing and you hope that you are right. In very long-term investment it's a good idea, but in trading not at all.
I trade all markets the same way , and it works, why isnt the EA working?
Optimizing for one cross will have its problems, thats where diversification could come handy, but what do i know lol

Ocsag Nov 18, 2018 10:03am | Post# 363

2 Attachment(s)
This is quite interesting. I am testing Feex edition with "trailing stop" as in post number 270. The only exception I did was to use TP55 and SL68 instead of the default TP68 & SL55 (i.e. swap the figures). Have a look at the results comparing both settings. It seems with a trailing stop/step using a lower TP and a higher SL yields better results.
During the test I have used a starting capital of 40873$ since this is the actual amount I have in my real account. Next week I will start trading with this EA.

Click to Enlarge

Name: TP68 SL55.jpg
Size: 202 KB

Click to Enlarge

Name: TP55 SL68.jpg
Size: 191 KB

mrdfx Nov 18, 2018 11:30am | Post# 364

Personally I wouldn't trust any back test data that is not 99.9%.

You can back test successfully on 90%, and then test again with 99.9%, and you will have vastly different results often times less successful.

Testing on 90% can give some idea of EA operation, but nothing conclusive and anything less than 90% is irrelevant.

tiborf71 Nov 18, 2018 11:39am | Post# 365

This is quite interesting. I am testing Feex edition with "trailing stop" as in post number 270. The only exception I did was to use TP55 and SL68 instead of the default TP68 & SL55 (i.e. swap the figures). Have a look at the results comparing both settings. It seems with a trailing stop/step using a lower TP and a higher SL yields better results. During the test I have used a starting capital of 40873$ since this is the actual amount I have in my real account. Next week I will start trading with this EA. {image} {image}
I've been testing it, and there are months, too big a loss.
I would not use live accounts.
but I may be wrong.

robots4me Nov 18, 2018 12:46pm | Post# 366

This is quite interesting. I am testing Feex edition with "trailing stop" as in post number 270
Your stats show a very short data horizon (approximately 2 months) and there are only 14 trades. Statistically this is not significant. Try using a longer data horizon (e.g. about a year) -- you really need about 100 trades for the stats to be meaningful. That's not to say your observation isn't real -- just that they are not statistically meaningful. If your observation is real then it will hold up using a longer data horizon.

For example -- suppose I choose a data horizon where there are 2 trades -- both winners. Does that mean the EA wins 100% of the time? Probably not. But suppose there were 100 trades -- all winners. In that case it could very well be it wins all the time. You need a large enough sample size -- e.g. 100 trades -- in order for the results to be statistically significant and taken seriously. That's the whole purpose of the Tester Report, right? Statistics (using historical data) is the only way we have to evaluate the potential performance of an EA -- other than actually trading it.

Ocsag Nov 18, 2018 1:24pm | Post# 367

{quote} suppose I choose a data horizon where there are 2 trades -- both winners. Does that mean the EA wins 100% of the time? Probably not. You need a large enough sample size -- e.g. 100 trades -- before you can take the results seriously.
Obviously you missed the point. I am not trying to prove the success or failure of the strategy. The idea was to compare two settings during the same period of time regardless whether short or long. And I was surprised to see that lower TP/higher SL with trailing stop gave far better performance than the default settings. This is an eye opener for me and I will continue carrying out more tests. In parallel I will put this EA to the real test with real money starting from tomorrow.

salimc Nov 18, 2018 2:15pm | Post# 368

{quote} Hi, Not only M1 can recalibrated, if we use this word. Only it is a matter of time needed to do the optimisation. I did also for real tick data. Took about 8 hours time for 10k passes. BR
Hi emonts,
your indicators are really impressive. keep it up!!!

robots4me Nov 18, 2018 2:31pm | Post# 369

Obviously you missed the point.
Actually -- no. You don't have a clue about statistics. This issue has been touched on before -- both here and elsewhere.

Correct -- this is not about whether the EA is good or not but, rather, how one goes about determining the best settings for an EA.

In order to compare two things (your new TP,SL settings versus the default TP,SL settings) you need to take a number of measurements. You took 14 measurements within a specific 2 month period and concluded that TP55, SL68 were better than the defaults for all time periods. That's simply not true. Taking 14 measurements or 5 measurements or 25 measurements during a different period might show the default values work better. Or they might show some other TP,SL combination work better. The data used for your testing will determine which TP, SL combination works best. If you use a different chunk of data, then you will get a different best TP, SL.

For example, using MT4's Strategy Tester and optimizer I found that during that same period of time a different TP,SL combination gave a profit factor of 8.62 and a win percent of 93.3% based on 15 trades. Those are much better than yours. What do you make of that?

This is where statistics and sample size comes in. It tells you whether your testing was sufficient to validate your conclusion. Taking 14 measurements (i.e. 14 trades) within a specific time period (2018.09.16 - 2018.11.14) is not sufficient to conclude the best settings. It only means that between 2018.09.16 - 2018.11.14 your settings worked better. It does not mean your settings will work better for other time periods or even in the future.

majamivice Nov 18, 2018 2:49pm | Post# 370

4 Attachment(s)
Here are some 99.90% BT:
Original EA from OP:
Click to Enlarge

Name: Screenshot_Original.png
Size: 72 KB

With exponential lot SL25 and SL55 and SL25Lot0.4:
Click to Enlarge

Name: Screenshot_exponential SL25.png
Size: 98 KB
Click to Enlarge

Name: Screenshot_exponential SL55.png
Size: 96 KB
Click to Enlarge

Name: Screenshot_exponential SL25 Lot0.4.png
Size: 99 KB

majamivice Nov 18, 2018 3:20pm | Post# 371

1 Attachment(s)
BT 90.0% the same as last above:
Click to Enlarge

Name: Screenshot_1.png
Size: 80 KB

mrdfx Nov 18, 2018 3:32pm | Post# 372

2 Attachment(s)
Here's 5 Years back tests of ULTRON with 99% modeling quality.

1st test is $1k start with 0.1 lots exponential.

2nd test is $1k start with 0.1 lots exponential using TS.

Raising the lot size to 0.2 increased the profits considerably, but also increased the DD so its up to you to know whats your risk tolerance if using higher lot size.

This EA passed last 5 years back test safely, and made a good stable profit, well done to the OP for creating this tool which we can add to our trading arsenal.
Click to Enlarge

Name: Screen Shot 2018-11-18 at 18.31.02.png
Size: 353 KB
Click to Enlarge

Name: Screen Shot 2018-11-18 at 18.31.28.png
Size: 359 KB

mrdfx Nov 18, 2018 3:43pm | Post# 373

{quote} take it {file}
Can you please share the source code for your version? Thanks.

meepstone Nov 18, 2018 3:44pm | Post# 374

2 Attachment(s)
Here are some 99.90% BT: Original EA from OP: {image} With exponential lot SL25 and SL55 and SL25Lot0.4: {image} {image} {image}
Using a 25 SL going back further than your 2018 start date results in worse performance.
Click to Enlarge

Name: 55 sl.png
Size: 63 KB
Click to Enlarge

Name: 25sl.png
Size: 63 KB

Spikehunter Nov 18, 2018 4:56pm | Post# 375

Hi all,
Having read most of the thread and looked at the code I have one question which I would like the answer to If there is a mathematician out there. Given 2 moving averages how do you calculate the maximum value spread that can occur between them. In the code this is set to 0.0048. Anything greater than this shows increasing momentum. This could be the key to determining the value for other pairs.

majamivice Nov 18, 2018 5:00pm | Post# 376

{quote} Using a 25 SL going back further than your 2018 start date results in worse performance. {image} {image}
But trading in 2013 and 2018 is completely different. We don't live in past.

meepstone Nov 18, 2018 5:44pm | Post# 377

2 Attachment(s)
{quote} But trading in 2013 and 2018 is completely different. We don't live in past.
Okay, starting from 2017. Only one year prior to yours still has worse results. What this is telling me is the 25 stop loss has more losses and if you had a wider SL like the original settings, it lets pullbacks happen and then still continue in same direction. Since in both cases the loses, including looking at your strategy tests show about double the amount of losing trades.
Click to Enlarge

Name: og.png
Size: 63 KB
Click to Enlarge

Name: 25slll.png
Size: 64 KB

robots4me Nov 18, 2018 7:36pm | Post# 378

Ladies and gentlemen...

With regards to which settings are best -- there are no setting values that work best for all dates. An EA is a simple algebraic formula -- if you don't believe me, just take a look at the *.mq4 source code (it is only a few lines) and see for yourself. Settings that work best will vary depending on the input data (i.e. the test period).

If you intend to trade in the present, then you test with recent data. The one caveat is your test period should include a minimum number of trades (e.g. 100) -- otherwise the results may not be statistically valid.

Built into the MT4 Strategy Tester is an "Optimization" tool. You really must learn how to use it. The "Optimizing" tool can be used to calibrate each setting. In this way you let MT4 compute the best settings for each test period. Forget about all these silly charts -- they are useless content and only serve to fill-up the web page. No one knows what to do with them -- not even the person who posted them.

Finally -- don't over-think this SL, TP issue. You don't understand the market well-enough to make sense of how they are affected by all the interdependent and dynamic variables at play. Very few people understand all the things that are in play -- I certainly don't. If you don't trust @Pawel's default settings, then the best you can do is to let MT4 empirically compute the best values.

Ansab001 Nov 19, 2018 3:02am | Post# 379

Ultron has set off a trade this morning for myself...

Eve55 Nov 19, 2018 3:05am | Post# 380

Ultron has set off a trade this morning for myself...
Yes to me too


© Forex Factory