Question for experienced trade system creators and theorists | 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

  • #1
  • First Post: Edited 5:44am Apr 20, 2013 12:21am | Edited 5:44am
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 checked.)


Each System:

- runs over the same time period (Result for the last 3 month)
- on the same 20 pairs
- each pair on 4 different TimeFrames

a Total of 80 different charts so to speak



Basically there are 4 Type: as some are very similar
- around 1050 trades in 3 month: Win % ca. 59 %
- around 520 trades in 3 month: Win % ca 64%
- around 220 trades in 3 month: Win % ca 65.3%
- around 108 trades in 3 month: Win % ca 63.9%



Attached Image
Money is important, Love is essential.
  • #2
  • Apr 20, 2013 4:06am Apr 20, 2013 4:06am
.
As a long time strategy creator it is impossible to give an opinion without knowing the strategy itself.
"You" are the only one that could work in the optimization efficiently. But here are a few general ideas
from experience:

First...If the strategy uses "any" type of indicators and if the results shown are "already"
optimized (curve fitted to the past data). Then, that is the end of it. You know what the results "would have been"
but irrelevant to a next/future period.

If the strategy "is not" 100% mechanical (uses any type of discretion from you) Then impossible to be
optimized by no one other then you.

If the strategy is 100% mechanical (to the point that if you run it with a code you get the same results)
but use some type of indicator, then the only thing you can do is to go back in history. Run, say 3 months
and curve fit the best possible... then use those parameters to run the next 3 months and see the results.
After those 3 month.. curve fit again with the latest 3 month and run the next 3... and so on.

If after going tru a year or so of data , you will know that is ok to run a "next period" with last period settings.
If you do not get good result running past data setting in future data, then you know the strategy itself is no good.

J.
  • #3
  • Apr 20, 2013 4:43am Apr 20, 2013 4:43am
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...

You have managed to make the question very vague? what exactly are you asking?

The No Brainers did a lot of work on optimizing their strats/entries... and in general, across the board, big R/R always made the most impact on results.
tradewith60
  • #4
  • Apr 20, 2013 5:48am Apr 20, 2013 5:48am
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?
Money is important, Love is essential.
  • #5
  • Apr 20, 2013 6:36am Apr 20, 2013 6:36am
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?
If you are working with a 1/1 R/R, and like you said, can adjust the lot size to obtain the same profits... the only thing that would justify
to go the 1050 trades route is if you are with a new Broker that give you a bonus according to volume traded.

J.
  • #6
  • Apr 20, 2013 6:42am Apr 20, 2013 6:42am
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?
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/
tradewith60
  • #7
  • Apr 20, 2013 6:46am Apr 20, 2013 6:46am
Much the same as Jeuro has said already.
Is this 3 months of backtesting or 3 months of testing in real time.
Lots of people do backtesting using a strategy tester and tweak the parameters until they achieve the best theoretical results.
This is curve fitting or to quote someone else, can't remember who - "Firing an arrow at a wall and then painting a target around it - Bullseye every time"
If this is a case of backtesting and curve fitting, then pick some other random 3 month periods and test again to see if you get similar results.
It is always best to forward test, ie run your EA or whatever on demo in real time for at least 3 months and see if it returns expected results.
You can forward test all 4 on demo and then pick which one works best, trade with that, but continue to run all 4 on demo and re-assess every 3 months.
Please Do Not PM Me With Coding Enquiries
  • #8
  • Apr 20, 2013 7:09am Apr 20, 2013 7:09am
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?
It really depends on the average length of time that you are in a trade, how many trades are you likely to have open at any time, especially with correlated pairs and what is the total risk to your account at any one time with multiple open trades. What is the anticipated losing streak, taking into account that there will probably be open trades with correlated pairs.

220 trades over a 3 month period is 3 or 4 per day, trades may or may not overlap and may or may not be with correlated pairs, we don't know with the limited information from you. If you are entering 15 or 20 trades a day with the other scenario, there is sure to be overlap and you will have a lot more open trades at any one time.

You have to decide how much you will be willing to risk per trade. Maybe you decide that you are prepared to risk 1% per trade when taking 220 trades over a 3 month period. A 65% hit rate will be giving you 20% profit per month. Most traders would be happy with that.

If you are entering 1050 trades over a 3 month period, you will have more overlapping or correlated trades and you should reduce the risk to take this into account. You will probably be better off with fewer trades, but with the confidence of a higher hit rate and able to confidently allow a higher risk per trade.
Please Do Not PM Me With Coding Enquiries
  • #9
  • Apr 20, 2013 11:19am Apr 20, 2013 11:19am
Hi!
I don't consider myself an experienced system creator but I'm writing one. You say you are closed to discuss the R:R of 1. I hope you'll change your mind by yourself...

Same for the 3 months test period. You need at least testing your system over
- An impulsive up trend
- An impulsive down trend
- A volatile up trend
- A volatile down trend
- A wide range
- A narrow range
- A choppy market
- A market with a central bank intervention

To me the best optimization option is to not optimize at all. As few parameters as possible. Market can change completly at any time. Things shall adapt themselves to the PA. The parameters shall not be critical at all: a big variation of the parameters values must end up with a small variation of the results.

Quoting Ati Baia
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%
Since the RR is the same for both, the system with 65% win/loss is better: the probability of a N-loser streak is smaller. => smaller DD => allows for a bigger lot size.

Also consider that you system's probability of winning is not what you think it is. Let's take the example of the system with 108 trades. You say it has a win% of 63.9%. I suppose you had 69 winners and 39 losers. But even with a 10% win rate you could end up with 69 winners out of 108. The probability is negligeable, sure. But it can happen. Actually your system's probability of winning is between 55.4% and 72.4% with a confidence of 90%. I assumed the trading decisions were independent. The image is the probability density function of the 69 winners out of 108 trades. Of course it peaks at 63.9%.
=> Adapt your MM on the worst case 55.4%.
Attached Image
No greed. No fear. Just maths.
  • #10
  • Apr 20, 2013 5:50pm Apr 20, 2013 5:50pm
I attach one of the Excel examples of all pairs: with some more info: Average Time, Max consecutive and some example running it on a 5000 start Capital
Attached File(s)
File Type: xls Result.xls   26 KB | 439 downloads
Money is important, Love is essential.
  • #11
  • Apr 20, 2013 5:58pm Apr 20, 2013 5:58pm
Thanks for all your feedback so far:

Quoting Gumrai
It really depends on the average length of time that you are in a trade, how many trades are you likely to have open at any time, especially with correlated pairs and what is the total risk to your account at any one time with multiple open trades.
Hi Gumrai,

- You are very right about the problem of correlated pairs / trades. Also using TF M5, M15, M30 H1 gives some additional correlated trades.

One reason I choose in my test 20 pairs and for each pair 4 TF is to avoid too much curve fitting. Only a couple of the 80 test charts where real failures.

The reason I only choose 3 month is quite simple: I'm with Oanda and I get for M5 min charts about so many data. I can download from Metatrade but it seems they do not offer all the pairs.

- your other point is also good to consider: multiple open trades which at this point I did not care for.
Money is important, Love is essential.
  • #12
  • Apr 20, 2013 6:05pm Apr 20, 2013 6:05pm
Quoting PipMeUp
Since the RR is the same for both, the system with 65% win/loss is better: the probability of a N-loser streak is smaller. => smaller DD => allows for a bigger lot size.
Hi PipMeUp, thanks for your response. This is one of the things I'm actually not completely sure of.

Because as I see it at the moment: more trades generally give more opportunities for money management: e.g. compounding ext..

Quoting PipMeUp
Actually your system's probability of winning is between 55.4% and 72.4% with a confidence of 90%.
I have to admit I don't understand it.
Money is important, Love is essential.
  • #13
  • Apr 20, 2013 6:08pm Apr 20, 2013 6:08pm
Quoting jeuro
If you are working with a 1/1 R/R, and like you said, can adjust the lot size to obtain the same profits... the only thing that would justify to go the 1050 trades route is if you are with a new Broker that give you a bonus according to volume traded. J.
Thanks. never thought about that maybe one can get something out of it...
Money is important, Love is essential.
  • #14
  • Apr 20, 2013 6:10pm Apr 20, 2013 6:10pm
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/
Thanks for the link.
Money is important, Love is essential.
  • #15
  • Apr 21, 2013 5:43am Apr 21, 2013 5:43am
here are 4 Result Excel sheets

RR:1:1

108 Trade Win 63.89% I guess not the best: better the next once

219 Trade Win 65.30%
531 Trade Win 64.40%
1084 Trade Win 59.41%


I always thought having more trades: (disregarding at the moment the problem of correlation an multiple trades at the same time) is in general an advantage (given a similar Win%) it allows smaller Risk per trade - the resulting curves are much smoother.

What is your opinions on that?
Attached File(s)
File Type: xls 108Trade_Win63.89Perc.xls   30 KB | 401 downloads
File Type: xls 219Trade_Win65.30Perc.xls   30 KB | 364 downloads
File Type: xls 531Trade_Win64.40Perc.xls   31 KB | 368 downloads
File Type: xls 1084Trade_Win59.41Perc.xls   31 KB | 371 downloads
Money is important, Love is essential.
  • #16
  • Apr 21, 2013 6:54am Apr 21, 2013 6:54am
.
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.
  • #17
  • Apr 21, 2013 6:58am Apr 21, 2013 6:58am
Quoting Ati Baia
more trades generally give more opportunities for money management: e.g. compounding ext.
The more trades in a given period the more likely they will be correlated. But I don't know you system. The entry prices will be very near one of the others. In short you get a group of winners or a group of losers. It becomes more or less equivalent to opening one single position with the risk of the group: over-leverage.

For the probability stuff. Let's use a very small example. Say you get this sequence: win-win-loss-win-win-loss-loss-loss-win-win. You want to know the probability of winning. The first approach is to say you have 10 trades and 6 winners. That makes 60% winners and a probability of 0.6. The complementary probability is 100%-60%=40% of losing.

Now, let's look at this sequence differently. Let's start from the probability. It's our unknown. Let's call it P. Either you win or you lose. The probability of losing is 1-P. I consider the outcomes of the trades are independent. With this probability P, the sequence above can appear with the probability
P x P x (1-P) x P x P x (1-P) x (1-P) x (1-P) x P x P.

Since A x B = B x A You can reorder this way:
P x P x P x P x P x P x (1-P) x (1-P) x (1-P) x (1-P). Which is P^W x (1-P)^L. W is the number of winners and L the number of losers.

Here P^6x(1-P)^4. If you draw this function you get a curve like the one I posted. It represents the probability of this sequence appearing for a given probability. It's the probability density function (http://en.wikipedia.org/wiki/Probabi...nsity_function) of your probability. It's only defined between 0 and 1 because your probability is somewhere between 0% and 100%.
Attached Image

The area below the curve is 1/2310 (integration of P^6x(1-p)^4 over [0,1]). The probability we search is very likely to be near of 0.6. Let's calculate the integral of the function between 0.6-d and 0.6+d, solving (numerically) for d such that the result covers 90% of the curve. You find d ~ 0.23. You can now say that you have 90% confidence that the probability is between 0.37 and 0.83. The estimation 60% comes with a big error! A system with R:R=1 is a winning system with a 60% win rate but it is a losing system if the win rate is 37%.

The more trades the narrower the curve is. You don't have to do those maths. The error is certainly way smaller for your 1080 trades sample. But take this one in excel. In column A you put 1 if the trade was a winner and 0 if it was a loser. In column B you put the sum of the previous 100 cells in A starting at row 100, (set B100 =SUM([A1:A100]) then copy-paste). You get a running win rate as a percentage. Plot the column B. It's certainly not flat.
No greed. No fear. Just maths.
  • #18
  • Apr 21, 2013 7:01am Apr 21, 2013 7:01am
Quoting Ati Baia
The reason I only choose 3 month is quite simple: I'm with Oanda and I get for M5 min charts about so many data. I can download from Metatrade but it seems they do not offer all the pairs.
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/
No greed. No fear. Just maths.
  • #19
  • Apr 21, 2013 7:14am Apr 21, 2013 7:14am
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.
Please Do Not PM Me With Coding Enquiries
  • #20
  • Apr 21, 2013 7:25am Apr 21, 2013 7:25am
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.
+1
That's exactly what I meant when I said
Quoting PipMeUp
To me the best optimization option is to not optimize at all. As few parameters as possible. Market can change completly at any time. Things shall adapt themselves to the PA. The parameters shall not be critical at all: a big variation of the parameters values must end up with a small variation of the results.
No greed. No fear. Just maths.
0 traders viewing now