Optimized Trend Trading | Page 20 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Optimized Risk vs Reward Equation 335 replies

Loss Less Optimized Strategy 63 replies

Optimized Parameters in Metatrader Testing? 3 replies

"Optimized Parameter" feature in EA? 0 replies

Strategy with Optimized parameter 1 reply

Attachments: Optimized Trend Trading
Exit Attachments

Optimized Trend Trading

  • #381
  • Feb 16, 2009 12:38am Feb 16, 2009 12:38am
Barhopper,

Sorry I was not trying to say that you should move to usual technical analysis, I was meaning that as you can't use this method directly to trading as curve, you need to find some specific ways of using this kind of techniques to trading.

Because the signal processing algos are eliminating noise and therefore making the past moves more simple, one real world trading implementation possibility might be to apply some shape recognition/statistical algorithms to the generated curves to approximate next direction or length.

The problem with statistical methods is that the amount of data would need to be quite large and that would probably restrict you to use lower timeframes that are (IMHO) quite unpredictable/unsuitable for trend style trading.

ps. to be honest, I'm not quite sure how this is so different from moving averages, the lag with moving averages is not really as critical as is the fact that this method repaints the whole curve at every point when new data arrives.
  • #382
  • Feb 16, 2009 12:39am Feb 16, 2009 12:39am
Hello,

thanks from my side too for all information included in this thread, especially during the last 2 weeks.
For those who are not so familar with the works of John Ehlers, I want to add the following link with a lot of ppt files:

http://www.mesasoftware.com/seminars.htm

Sixer
  • #383
  • Feb 16, 2009 9:51am Feb 16, 2009 9:51am
Hi all,

This is a question that comes up all the time with intraday timeframes... If the bar period is short and/or the trading is thin, there will be missing bars from time to time. I wonder what your attitudes are in dealing with them.

Many of the techniques we've discussed here assume evenly-spaced samples, whatever that means. They could be evenly spaced in time (conventional interpretation) or evenly spaced in transaction volume, or evenly spaced in transaction count. In the latter two cases there will be no missing bars, by definition.

But assume we are using time-based bars, and there are bars for which no transaction occurred. Does anyone have a heuristic for when it's best to

1) Fill the missing bars with the last known price
2) Pretend the bars aren't missing, and just always count N bars back
3) Vary N as necessary to ensure your sliding window is always T minutes wide, without filling in any bars within that window
4) Do something else?

The second approach is often easiest, and I suspect it's what most people do (consciously or not). And it may not matter all that much. But if you think it does I'm curious to know your thoughts. Thanks...
  • #384
  • Edited Feb 17, 2009 9:20am Feb 16, 2009 10:37pm | Edited Feb 17, 2009 9:20am
Thanks for the link Sixer... Those were good presentations. However I did find a glaring error (both conceptual and mechanical) in Ehlers' slide on Bertrand's Ballot Theorem. Did anyone else see that?

But the point he makes on that slide is directionally correct... every strategy has a very real chance of going underwater. Out of curiosity, I ran a quick Monte Carlo simulation of the probability of staying above water (equity curve staying > 0) over a period of 100 trades, given a win rate and a per-trade profit factor (i.e. there's a fixed profitTarget and stopLoss at the time of entry).

Caveats: 1) These assume a purely random ordering of winning/losing trades. Trade series that are correlated in time (wins tend to follow wins) or anti-correlated (losses tend to follow wins) will yield different results. 2) I'm assuming no money management here... underwater simply means the equity curve went below 0 at some point in the trade series.

I'm attaching an image, and the Matlab figure file in case anyone wants to click it and rotate it, etc. On the vertical line where profitFactor==1 it reduces to Bertrand's theorem, but only on that line. You can have a 90% chance of staying above water even if your profit factor is only 0.5, but you have to have a winRate of 97%. Not reasonable by most standards!

On the other hand, if you have a profitFactor of 2.0 and a winRate of 60%, you'd think you had a good system, right? Well you'll have a 50% chance of losing all your winnings (equity curve going underwater) if you keep trading it long enough.

I'll end on that happy note Cheers...
Attached Images
Attached File(s)
File Type: zip probAboveWater2.zip   26 KB | 618 downloads
  • #385
  • Feb 17, 2009 1:59am Feb 17, 2009 1:59am
Hello,

I want to share another link which contains a great eBook collection for Matlab users:

http://www.4shared.com/dir/8284699/7...ollection.html

Sixer
  • #386
  • Feb 17, 2009 6:21am Feb 17, 2009 6:21am
Quoting BarHopper
Thanks for the link Sixer......
Hi,
I've goit a question:
Do you understand "profit factor " as the factor of money won i a winning trade compared to a lost money in a losing trade? For example if p.f.=2, then when we win, we win $2, and when we lose a trade, we lose $1?
Do you understand "win rate" as a percentage of winning trades out of all trades?

If yes, I am almost sure, that your calculations are maybe correct, but are made with a very bad money management. Please try doing 1000 trades, but set that each trade is risking no more than 5% money you had when you were beginning (of 5% of the current money, if you have more).
And attach your .m file
  • #387
  • Feb 17, 2009 10:00am Feb 17, 2009 10:00am
Hi Julek,

Yes, you are correct, those simulations were done with no money management whatsoever... I've edited my post to clarify this. By "underwater," I'm not saying that you'll lose all your capital, just that you'll lose all your profits. If you stop there, you still have 100% of your capital.

I suppose an optimist could look at the previous plots and say that with a profitFactor of 2 and a winRate of 60% you've got a 50% chance of making it big even if you start with minimal capital, and leverage yourself like crazy. But that optimist would be a gambler

So what about money management? The attached slide shows the answer to your specific question. I did 1000 trades per trial and changed the definition of "above water" from min(equity)>0 to min(equity)>-20, since with a 5% bet you'd have to lose 20x your bet size over the life of your tradeseries to wipe out all your capital. As you can see, the likelihood of going underwater now (under this new definition) is much lower, and is almost zero provided the winRate is a bit above 1/(1+profitFactor), which is the breakeven winRate.

So this is more reason to give money management the attention it deserves. But I've gotten off topic... if we want to start a money management thread (or post on someone else's) then great, but I don't want to derail Codebreaker's original topic. Cheers...
Attached Image
  • #388
  • Feb 17, 2009 12:19pm Feb 17, 2009 12:19pm
Quoting BarHopper
since with a 5% bet you'd have to lose 20x your bet size over the life of your tradeseries to wipe out all your capital.
huh?

Are you assuming that the betsize is calculated from the original deposit? Of course it is not. The bet size must always be calculated from current account balance.
  • #389
  • Feb 17, 2009 3:20pm Feb 17, 2009 3:20pm
Quoting Codebreaker
Hi Challenger

You are absolutely right. Some timeframes are definitely better than others for different pairs. ...

...Sadly, MT only allows 9 different timeframes. In Matlab, I collect the tick-data via DDE and can display any TF I want. I currently like the 11min 43.2sec TF on the EURGBP!! I have just done 19 consecutive scalp trades on this with no loser.

CB
Hope I'm not too late in replying to a post from twelve days ago. I'm a long time lurker--but this is a great thread, so here goes....

There is a way to create graphs in MetaTrader in nonstandard time frames such as M2, M7 etc. I found this on another FF thread. Create a graph in a standard time frame and add the indicator Period_Converter_Opt to it (I know 1 Minute graphs work and probably others as well).

Then click on Flie/Open Offline and select your time frame. Make sure the new graph is updating and keep the first graph open or minimized.
Attached File(s)
File Type: mq4 Period_Converter_Opt.mq4   15 KB | 702 downloads
File Type: mq4 Period_Converter_Opt2.mq4   15 KB | 673 downloads
  • #390
  • Edited 4:17pm Feb 17, 2009 3:42pm | Edited 4:17pm
Quoting mikkom
Are you assuming that the betsize is calculated from the original deposit? Of course it is not. The bet size must always be calculated from current account balance.
Yes mikkom, I assumed the bet size was static based on 5% of the initial deposit. Would you do that in practice? Maybe, if you made a practice of frequently withdrawing or paying out your profits, as an "income investor" might prefer. But if capital appreciation is your goal, you would probably reinvest your profits, and the bet size would increase/decrease along with your account.

If you would prefer to see it the other way, you are free to model it. (Actually, if you always adjust your bet to be 5% of your current capital, there is no way to be wiped out, and the probAboveWater surface is then flat at 1.0.) Anyway, we're getting off the point of this thread. If there is a money management thread somewhere and you want to continue this conversation over there, please PM me.

BTW, does anyone have an opinion on my post #383? I think it's fairly relevant to the earlier discussion. Cheers...
  • #391
  • Feb 17, 2009 7:32pm Feb 17, 2009 7:32pm
Quoting BarHopper
Hi all,

This is a question that comes up all the time with intraday timeframes......
Hi BarHopper,

I may be derailing the purpose of your questions slightly, but I see another place where this question comes into play, and that's where you don't have 24 hour markets as we essentially do in FX.

I usually fill in missing data with the most recently available price. This is probably fine for daily data, but when it comes to intraday data, I'm not sure.

I've beening thinking most recently about the case where you might want to look at equity, bond or commodity contracts that may only trade for an 8 hour day and then have 16 hours of silence. What is the best way to deal with the transition from today's closing price to tomorrow's open?

Is it to pretend that it's no different from any other bar?

I don't think so, but I also don't have an appropriate alternative.

Any thoughts?

Thanks,
Ngawang
  • #392
  • Feb 17, 2009 7:34pm Feb 17, 2009 7:34pm
Quoting BarHopper
Hi all,

This is a question that comes up all the time with intraday timeframes... If the bar period is short and/or the trading is thin, there will be missing bars from time to time. I wonder what your attitudes are in dealing with them.

Many of the techniques we've discussed here assume evenly-spaced samples, whatever that means. They could be evenly spaced in time (conventional interpretation) or evenly spaced in transaction volume, or evenly spaced in transaction count. In the latter two cases there will be no missing bars, by definition....


Hi Barhopper,

I messed up your original quote when writing that last post. It was probably clear, but just in case, this is the post I was referring to...

Thanks,
Ngawang
  • #393
  • Edited 3:54am Feb 18, 2009 2:32am | Edited 3:54am
Quote
Actually, if you always adjust your bet to be 5% of your current capital, there is no way to be wiped out, and the probAboveWater surface is then flat at 1.0.
Exactly my point.

Quoting BarHopper
Anyway, we're getting off the point of this thread. If there is a money management thread somewhere and you want to continue this conversation over there, please PM me.
I agree. However I just pointed out errors in your calculation - I think it's quite important because otherwise people tend to believe professional looking graphs like the one you posted.

I tried to suggest some possible methods of exploiting smoothed curves that are produced by these smoothign techniques, anyone read those? As I said before, I'm interested in practical approaches.

One practical approach could be that you would extract basic shapes of the curve produced by signal processing [peaks, heights between them etc] on training period and filter them through a bayesian filter to produce up/down probabilities. Then verify that the results agree on testing period [which must be totally different of course].

If the results agree, you have just invented a practical method of using the curves produced to trading. [and if not, the results will tell you something about how good these curves are on predicting the future].
  • #394
  • Feb 18, 2009 4:29am Feb 18, 2009 4:29am
Quoting PiedPipper
Hope I'm not too late in replying to a post from twelve days ago. I'm a long time lurker--but this is a great thread, so here goes....

There is a way to create graphs in MetaTrader in nonstandard time frames such as M2, M7 etc. I found this on another FF thread. Create a graph in a standard time frame and add the indicator Period_Converter_Opt to it (I know 1 Minute graphs work and probably others as well).

Then click on Flie/Open Offline and select your time frame. Make sure the new graph is updating and keep the first graph open or minimized....
PiedPiper

You are very welcome indeed on my thread and thanks for "lurking"! Your post is extremely helpful!! I am very excited that we can now get non-standard time-frmes in MT - I had no idea this was possible. This could be a very powerful tool.

(And thanks for getting the thread back on track - all this talk about MM is nothing to do with our topic...)

CB
  • #395
  • Feb 18, 2009 11:02am Feb 18, 2009 11:02am
CB-- You're very welcome. Glad to be of help. Now you can scalp on 11 or 12 minute MT bars at will!

--Pipper
Attached File(s)
File Type: mq4 Period_Converter_Opt3.mq4   15 KB | 851 downloads
  • #396
  • Feb 18, 2009 11:54am Feb 18, 2009 11:54am
There have been a few posts by people objecting to using non-causal filtering or non-causal methods of any kind because they assume that you have that look into the future when filtering past data. This ultimately gives the 'repainting' on the most recently filtered data that we clearly don't want in our process.

It seems like the justification for using non-causal methods has to be an assumption that the cycles are inherently stable once noise has been removed in an optimal or at least reasonably efficient manner. (The word optimal is interesting and probably deserves a discussion on its own -- for instance, Kalman filters are labeled as optimal -- but optimal with respect to what? They don't seem to give optimal noise reduction for cycle measurement.)

So if that's the case, then we should be ok using non-causal methods for noise reduction.

So on to the next question. If this is valid, and methods such as HP filtering and SSA are now useful tools for removing noise before measuring cycles...

Why not use something very simple like the Exponential Moving Average Filter?

This is heresy -- right? We've already discussed how useless it is because it gives a terrible phase response. But what if we filter both forwards and backwards through the data set?

I haven't tried this yet, but I suspect that if you filter forwards through the data set with an EMA, get the ugly phase output but then subsequently filter backwards through the data with the same filter parameters...

Wouldn't you remove the ugliness?

Is it worth a try?

Thanks,
Ngawang
  • #397
  • Feb 18, 2009 2:28pm Feb 18, 2009 2:28pm
Quoting ngawang
There have been a few posts by people objecting to using non-causal filtering or non-causal methods of any kind because they assume that you have that look into the future when filtering past data. This ultimately gives the 'repainting' on the most recently filtered data that we clearly don't want in our process.

It seems like the justification for using non-causal methods has to be an assumption that the cycles are inherently stable once noise has been removed in an optimal or at least reasonably efficient manner. (The word optimal is interesting...
Ngawang - that is a very cunning idea! It would certainly be MUCH simpler than programming Goertzel, Hodrick-Prescott, SSA, etc which are rather tedious in MT language.

Perhaps one of the group can give this a go and report back.

CB
  • #398
  • Feb 18, 2009 3:54pm Feb 18, 2009 3:54pm
Quoting PiedPipper
CB-- You're very welcome. Glad to be of help. Now you can scalp on 11 or 12 minute MT bars at will!

--Pipper
what is the difference between the Period Converter 1,2 and 3?

thanks for sharing.
  • #399
  • Feb 18, 2009 4:10pm Feb 18, 2009 4:10pm
Quoting giraia_br
what is the difference between the Period Converter 1,2 and 3?

thanks for sharing.

I think if you have Period Converter_1 on more than one graph it stops updating quotes on that chart. So 1, 2,3 allows you to run up to three graphs with non-standard time frames. You could probably create a Converter_4 and 5 to run more.
  • #400
  • Feb 18, 2009 9:18pm Feb 18, 2009 9:18pm
Hey CB,

Got an equity curve to show on TopCat? The equity curve should be pretty long by now. Over a year long!
0 traders viewing now