Question for experienced trade system creators and theorists | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Question for experienced Fibonacci traders 3 replies

Question for experienced traders: what am I doing wrong? 19 replies

Question for experienced MT4 Traders. 3 replies

Expectancy question for experienced traders 2 replies

Question for experienced traders 0 replies

Attachments: Question for experienced trade system creators and theorists
Exit Attachments

Question for experienced trade system creators and theorists

  • #21
  • Apr 21, 2013 1:02pm Apr 21, 2013 1:02pm
kudos @PipMeUp, couldn't have said it better.

I also recommended reading Curtis faith book Way of the turtle. He shows how one should optimize correctly ....brilliant book.
wo-yoy! wo-yoy! wo-yoy! wo-yoi! wo-yoy-yoy-yoy!
  • #22
  • Apr 22, 2013 6:24am Apr 22, 2013 6:24am
Quoting PipMeUp
{quote} Don't limit yourself for bad reasons. You can download histo data from here for instance https://www.dukascopy.com/swiss/lv/m...ch/historical/
hi PipMeUp,

Thanks a lot. Obviously your math brain is much more developed than mine

Thanks a lot for the Histo data link. I will have to look into it if there is a way to import them into Metatrade 4.

Very kind of you.
Money is important, Love is essential.
  • #23
  • Apr 22, 2013 6:33am Apr 22, 2013 6:33am
Quoting Gumrai
The question that still not been answered. Did you curve fit, ie adjust the parameters to give the best results from the set of data that you were using or did you decide on a strategy and then backtest it. If curve fitted then you are unlikely to achieve similar results in future unless tested over a much larger period than 3 months.
Not much curve fitting done. But of course I played a bit around with different SL distances and implementations. Anyway I will adjust my backtest code model a bit to better implement the suggestion of

60minuteman
Quoting 60minuteman
theres not enough information to make a decision there, you need to build a proper risk model for each strategy.. http://fxevolve.com/blog/how-to-buil...ng-risk-model/
I had already implemented most of it as I base most of my calculations on: % of SL Risk. So I will do the small missing parts and than try to run is on much longer data sets from if I can find a way to import the dukascopy data into MT4.


Thanks a lot to all of you..
Money is important, Love is essential.
  • #24
  • Apr 22, 2013 7:24am Apr 22, 2013 7:24am
I would say there is a greater margin for error in your results in the systems with the fewer number of entries recorded. So if you are to rely on the % wins figure then the systems with the largest sample (400+ trades) will be more accurate.

The other consideration as touched on with mention of correlated pairs is spread. 20 trades a day = you pay 20x spread, I can't quite work it out but would you not be better off discarding correlated trades and having fewer entries per day? But with double lot sizes and only 10trades per day I think you would pay the same anyway?
  • #25
  • Apr 22, 2013 8:52am Apr 22, 2013 8:52am
Quoting jeuro
. Ati baja. There one thing I do not really understand from you first post. You want to Optimize??? or you want to choose the best?? Obviously you are not a newbie, but why would you like to choose or optimize?. Even the worse stat you posted in your first post would be a holly grail. Divide your capital and trade them all. I do not see your dilemma or Am I missing something?? J.
Not a big dilemma but I don't believe that I can count an the same results when live trading. So I think choosing my best option/even improving it (without destroying it) would be only a wise thing to do.

Anyway, I never used MT4 for automatized live trading so I expect some additional issues too.

e.g. I do not mind to much about news as it will go sometimes in my direction and sometimes against me but there is a big related problem:

- e.g. Oanda increases at times the spread quite big time: which is better manageable when I trade manual but fully automatized this will wipe out quite a number of my trades: at least I guess so. will still work on some solution - I have a fully configurable Time based NewOrder execution blocker - but a trade once opened is at the mercy of such.

- Entry spreads do also vary a lot: I use some filters to block New Orders in such cases: but this will alter my result too -- probably less live trades than the test run. (of course I could be lucky and the bad once are skipped

- The other thing is: correlation and multiple similar trades opened: which I have to handle somehow not sure yet.
In my code I have this fully customizable but I still have to make a decision about it.

for my test runs which I always run separately for one Symbol/TFI at the time: I lock execution till the previous trade is closed. meaning there should be no overlapping trades on the same TF/Symbol.

- There are other issues which will alter my results like: Internet connection failures ect. ect.
- taxation


If the automatized way works out, whatever little profit is left in the very end is at least not much "work"




Anyway, there are already some helpful input I take on board and I'm grateful for each one who contributed so far.
Money is important, Love is essential.
  • #26
  • Edited 7:34am Apr 24, 2013 7:15am | Edited 7:34am
Quoting PipMeUp
{quote} Don't limit yourself for bad reasons. You can download histo data from here for instance https://www.dukascopy.com/swiss/lv/m...ch/historical/

A bit off-topic: just for anybody who might read this in the time to come:

-M1 historical data:
http://www.fxdd.com/mt/en/forex-reso...1-minute-data/

-and a Period converter might be also handy:
http://codebase.mql4.com/949
http://www.forexfactory.com/showthread.php?t=206301
Money is important, Love is essential.
  • #27
  • Apr 24, 2013 9:06am Apr 24, 2013 9:06am
did some small changes on the period converter: http://codebase.mql4.com/949

maybe someone finds it useful:
//+------------------------------------------------------------------+
//| 20130424: ATI: changed
//| - filename to external input parameter
//| - iTimeShiftHours to external input parameter
//| - added some Comments and FINISH Alert
//+------------------------------------------------------------------+


This should be the End of the off-topic - probably the end of the thread too
Attached File(s)
File Type: zip periodgen_changed_20130424.zip   3 KB | 260 downloads
Money is important, Love is essential.
  • #28
  • Apr 24, 2013 11:55pm Apr 24, 2013 11:55pm
in order to get 1 to 1 risk reward and to include the spread you have to have a larger reward on screen. if the math is wright for your systems than you first have to do the monte carlo analysis for each system to see which one is more likely to fail in the long run. if after 10000 simultations the one with the highest frequency does not produce failure i would go with that one. the higher the number of trades the better the results will be first of all in terms of probability and the law of large numbers and second in terms of profit. but you have to make sure that your system performs well in monte carlo simulations.

ati baia . i will trade with oanda also in the fututre on a manual basis because i don't know how to program at the moment . i want to ask you about the spread thing. wouldn't an expert see that the spread has become huge and immediately cancel the stop loss or move it further apart? my manual solution is to see the highest impact news and cancel the stops around that time and after that putting the stops back in. this is the only solution. but i wonder if a metatrader expert can see the spread in real time and cancel the stop loss.
  • #29
  • Apr 25, 2013 1:25pm Apr 25, 2013 1:25pm
hi deltatrade,

Thanks for the feedback: I tested the whole thing on 5 years and the results for each year is very much in line.

Only I'm not sure how to handle / account for Max consecutive Losses: they are pretty low but what I'm more worried about is: if I run multiple pairs/TF I'm not sure how to correctly account for them: If they happen at the same time - this would mean I have to cut down my initial Risk a lot - the other idea is of course to not allow too many trades at the same time - which I only have to configure.

never heard about the "monte carlo analysis" will have to check that out


Ati response in Blue;
quote=deltatrade:
i ati baia . i will trade with oanda also in the fututre on a manual basis because i don't know how to program at the moment . i want to ask you about the spread thing. wouldn't an expert see that the spread has become huge and immediately cancel the stop loss or move it further apart?

It all depends how such expert is coded - so yes and no

my manual solution is to see the highest impact news and cancel the stops around that time and after that putting the stops back in. this is the only solution. but i wonder if a metatrader expert can see the spread in real time and cancel the stop loss.

Like I said It all depends how such expert is coded - so yes and no

the problem I have with it is this: I do not want to move my SL / and the only way I move it is in profit direction to secure some profit/reduce possible loss.
I calculate my Risk, Position Size ect all based on the Initial SL: so if I move it further away my position size is for the potential loss too big.

Actually I'm quite satisfied with Oanda
even though I think their new change to variable spreads was a very bad decision (speaking as a client) most of the time the spread is now much bigger than it used to be some time back and it has become a bit more unpredictable even during normal trading time.
Money is important, Love is essential.
  • #30
  • Apr 25, 2013 1:56pm Apr 25, 2013 1:56pm
Quoting Ati Baia
Thanks for all your feedback so far: Yes I'm talking about 100% automatized trading the main question is: how to choose the best (what to consider) from successful systems based on the trades frequency and results. is it better( and why does one think so) to have - 220 trades with a Win of ca 65% - or 1050 trades with Win of 59% in both cases one could adjust the final result to be equal by adjusting the traded lot size accordingly: but what would you choose and why?
Maybe this helps?
http://www.investopedia.com/articles.../04/091504.asp


I am not sure, I fully understand the question, but using Kelly`s formula, it is much better to have more trades, your account grows faster.
  • #31
  • Apr 26, 2013 10:14am Apr 26, 2013 10:14am
I decided to keep some notes how my automated MT4 system develops: Maybe once in a while I get some useful feedback.


-----

At present:

START REAL TRADING:
Current decision:
Once I start real trading I will use in the beginning a higher Win% with less trades. (additional Filter configuration)

Even though this will cut my potential profit a lot: considering additionally compound profit
I do not plan to watch the screen 24/7 so I like to have an additional piece of mind knowing I have a System with relatively high Win% (which also means theoretically less max consecutive looser to care for)

Once this worked out: and a certain amount of profit is earned I might consider looking into increasing the number of trades accepting a lower Win% (I'm still convinced that certain kind of money management is easier to apply to a higher number of trades)


SPREAD:
- my original Test runs I did without any Spread parameters - just to see how the systems would perform in an optimal situation
- well, even 1-3 pips Spread cuts the Win % quite a bit

Current decision:
Test all with a 3 pips spread: which I will use as the maximum allowed spread (for any pair) to open a new Order


PARAMETERS:
Current decision:
Try to keep most parameters dynamic: e.g. don't use a fix SL pips setting
Reason: optimal there should be no need to adjust anything for different TF / Symbols - even though they behave quite different

Exceptions:
- Max Spread: FIX 3 pips any pair
- MIN SL: 11.7 pips any pair (if I accept in worst case 3 pips Spread I need some additional pips before my SL is hit)
- no MAX SL just calculated it each time dynamically


PAIRS:
Current decision:
Reduce for the test the 20 pairs to only the once I could download Data from: http://www.fxdd.com/mt/en/forex-reso...1-minute-data/

which finally gives me: 17 pairs on each on 4 TF


TF-CORRELATION:
I did some manual checks of a couple of TF and with some the less frequent systems they did not overlap to often - which was very good to see.

Current decision: Do not allow simultaneous trades of the same pair (about correlation between different pairs I still have to see)


RISK:REWARD %:
Current decision: stick for the start with 1:1

**** NEXT STEPS ***

BACKTEST and ADJUST

All 17Pairs/4TF: with 3 PIPS Spread

- yearly basis
- monthly basis
(skip the weekly option because of too less trades per pair)


Adjust Entry Parameters to produce at least an average of Win % 55


CHECK OUT: MonteCarlo Simulation 2.0 if I find it helpful for my situation

deltatrade kindly PM his MonteCarlo Simulation 2.0 link http://www.forexfactory.com/showthread.php?t=325567


AFTER THAT: Re-ASSESSMENT
Money is important, Love is essential.
  • #32
  • Apr 26, 2013 3:05pm Apr 26, 2013 3:05pm
My current backtest process:

I do not use MT4 Strategy Tester but have implemented my own small Test suit as an Indicator.
I call them: PseudoTrades

I run only rough tests: no tick data but Entry - and check if a bar hits SL or TP first.
In case the same Bar touches both I count such trade as skipped (unknown outcome)

I find this good enough and pretty fast.



1.
I use 1 year data and Run tests with the main parameters so far to the extrem that it actually starts to produce less favourable results
This are the boundaries

2.
Then I run 1 year test with the parameter changing step wise from I extreme to the other.

3.
get rid of all which failed my requirement of Min Win %: 55

4.
for the remaining: I run yearly test over 5 years of data

5.
if they confirm the positive results (similar Win% and similar number of Trades) I will select some and run also Monthly tests


===== Screenshot: I thought maybe it might inspire/help some other coders/traders

Here a screenshot of what I do for visual PseudoTrades backtesting.
Attached Image (click to enlarge)
Click to Enlarge Name: pseudotradebacktest.png Size: 35 KB
Money is important, Love is essential.
  • #33
  • Apr 27, 2013 8:34am Apr 27, 2013 8:34am
My planned approach from above did not work out as planned: so I'm doing it now differently.

will run all tests over the whole 5 years and over each year separately. (had to code the yearly split option)
Money is important, Love is essential.
  • #34
  • Apr 27, 2013 4:37pm Apr 27, 2013 4:37pm
Decided to code an own Monte Carlo Simulation for my integrated MT4 system.

for an Excel version see: deltatrade's : http://www.forexfactory.com/showthread.php?t=325567
there is a Version 3 at Post# 11
Money is important, Love is essential.
  • #35
  • Apr 27, 2013 5:21pm Apr 27, 2013 5:21pm
Quoting Ati Baia
To all experienced trade system creators and theorist. I hope to get some helpful input how to optimize / choose the best option from a couple of systems. I'm not looking for a particular system discussion or anything like that but how to choose the best (what to consider) from successful systems based on the trades frequency and results. Each Trade has a Risk/Reward Factor of 1:1 (also not looking to discuss this - I just mention it as a given factor) Time period is last 3 month (not looking for discussion how long backwards a system should be...
i will pick the one with 220 trades i think it will be more safe and have a bigger wining ratio in long term
if you need more safety pick the one with 520 trades

but i am not so experienced
have a nice day
focus while relaxed to see better
  • #36
  • Edited Apr 30, 2013 5:54am Apr 29, 2013 5:39pm | Edited Apr 30, 2013 5:54am
update: 2013.04.30: in case someone ever wonders: in the attached file there are still errors: the Max.Consecutive LOSS are not correct. Found a problem.


Quoting Ati Baia
Decided to code an own Monte Carlo Simulation for my integrated MT4 system. for an Excel version see: deltatrade's : http://www.forexfactory.com/showthread.php?t=325567 there is a Version 3 at Post# 11




Monte Carlo Simulation for MT4

This is what I came up with: maybe it inspires some others or someone has a good idea to improve it

(NOTE: this is making use a lot of my integrated code so I want post the MT4 script: would need quite some work.)

I based it on deltatrade's excel idea with some additions ect..


The script uses this input parameters:

//===TradeSequenceInput (take it from: e.g. an original Backtest / Live trading period)

i.Num.Total.Trades '1421'
i.Num.Winning.Trades '856'
d.Win.RMultiple '1.0' the Average Risk Multiple for Winning Trades (e.g. R:R 1:1)

//===MonteCarloSimulation
i.MonteCarlo.Num.Loops '5000' Number of MonteCarlo Test Sequences

each MonteCarlo Simulation Sequence uses 1000 random trades which allows for a Win% of one decimal point


Output is like the attached file:

Main output: all based on all Sequences & the whole sequence of where it occurred

Highest.Win %:
Lowest.Win %:
Highest.Final.RMultiple:
(Each sequence runs 1000 random trades but it also dependes on d.Win.RMultiple parameter a LOSS is always counted as 1 RMultiple)
Lowest.Final.RMultiple: (Each sequence runs 1000 random trades but it also dependes on d.Win.RMultiple parameter a LOSS is always counted as 1 RMultiple)
MaxRunUp.RMultiple: The Maximum Run Up over 1000 trades in RMultiple (Highest Point)
MaxRunDn.RMultiple: The Maximum Run DN over 1000 trades in RMultiple (Max DrawDown)
Max.Consec.WIN: The Maximum Number of Consecutive W Trades over 1000 trades
Max.Consec.LOSS: The Maximum Number of Consecutive W Trades over 1000 trades
Attached File(s)
File Type: txt MonteCarloSimulation_LOG.txt   9 KB | 275 downloads
Money is important, Love is essential.
  • #37
  • Apr 29, 2013 5:41pm Apr 29, 2013 5:41pm
Monte Carlo Simulation: Confidence Level Matrix:

Is the percentage over all MonteCarlo Sequences:

e.g. in the Post above attached file:

WIN %Column: Above 60% MonteCarlo Confidence %: 53.70

means: of the 5000 MonteCarlo Simulation Sequences: 53.7% ended with a Win % above 60%

in front of the same row under: Average: 60.19 means the average of the 5000 simulation sequences Win % was 60.19%



I guess I leave it with that - should be good enough for the purpose: except someone has a good input.
Money is important, Love is essential.
  • #38
  • Apr 29, 2013 9:34pm Apr 29, 2013 9:34pm
You also need up win% and dn win%, the best system will be the most balanced, the up and dn algorithm can handle the other side market fairly well.
  • #39
  • Edited 4:00am Apr 30, 2013 3:33am | Edited 4:00am
Quoting jgray110
You also need up win% and dn win%, the best system will be the most balanced, the up and dn algorithm can handle the other side market fairly well.
Thanks JGray,
You also need up win% and dn win%: Not exactly sure what you mean: Max DrawDown in %


I settled for MaxRunDn.RMultiple: which of course can easily be converted to Account % - Depending What one Risks per Trade (RMultiple)

If you meant something else please let me know

Cheers
Ati
Money is important, Love is essential.
  • #40
  • Apr 30, 2013 11:03am Apr 30, 2013 11:03am
PROCEEDING with my system:

* I like the Monte Carlo Simulation thing - because of this I will proceed like this.


1.
Define my Lowest acceptable WIN% (using R:R 1:1) considering the "Worst case" of:

200000 Monte Carlo Simulation:
- each sequence of 1000 random trade outcomes

Well, if my actual results are worse I will consider myself quite "unlucky"


2.
After that I will try to see if any of my systems can reach at least such return by acceptable trade frequency over the time period (2007 - 2012)
- R:R 1:1
- Max Spread: FIX 3 pips any pair/any TF
- MIN SL: 11.7 pips any pair/any TF

main issue I see at the moment is: the Simulations are based on 1000 trades - will I find a system which generates enough trades in a reasonable time. probably I will need to compromise here.

the other thing is how simultaneous trades might effect my max DrawDown at any given point in time
Money is important, Love is essential.
0 traders viewing now