Systematic trading | Page 18 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Trade another day through Systematic Trading 203 replies

Nonparametric Association Measures in Systematic Trading 57 replies

Anyone use Systematic from Alpari? 1 reply

A Systematic Approach to Markets 68 replies

Systematic Trading vs. Discretionary Trading 14 replies

Attachments: Systematic trading
Exit Attachments

Systematic trading

  • #341
  • Sep 1, 2009 4:35pm Sep 1, 2009 4:35pm
How are you selecting the final algo? is it the fittest of the final generation?
The breaking of a wave cannot explain the whole sea.
  • #342
  • Sep 1, 2009 5:00pm Sep 1, 2009 5:00pm
Quoting Craig
How are you selecting the final algo? is it the fittest of the final generation?
I don't have any final algos, the framework will generate them forever if I let the computer running, the pics I have posted are just shots from random points of the process. The selection process is actually very simple at the moment, I only look at profit factor for the in-sample period.
  • #343
  • Sep 1, 2009 5:58pm Sep 1, 2009 5:58pm
Are you adding in commission, slippage etc?
The breaking of a wave cannot explain the whole sea.
  • #344
  • Edited 1:59am Sep 2, 2009 1:26am | Edited 1:59am
Quoting Craig
Are you adding in commission, slippage etc?
I use bucketshop spreads for all pairs so commission is there and the real world commission will probably be cheaper, I don't add slippage unless the candle explicitly shows slippage (open above stop/below sl for example). I probably should though.
  • #345
  • Sep 2, 2009 5:27pm Sep 2, 2009 5:27pm
Quoting mikkom
I use bucketshop spreads for all pairs so commission is there and the real world commission will probably be cheaper, I don't add slippage unless the candle explicitly shows slippage (open above stop/below sl for example). I probably should though.
So you only open trades on the open (or close) of each candle (and also evaluate stops at the same point I assume)?
The breaking of a wave cannot explain the whole sea.
  • #346
  • Edited 2:40am Sep 3, 2009 2:18am | Edited 2:40am
Quoting Craig
So you only open trades on the open (or close) of each candle (and also evaluate stops at the same point I assume)?
I simulate the trade opening/sl/tp per 1 min candles - I simulate everything based on candle highs/lows (+spread). I simulate everything as pessimistic as possible, ie. if a candle hits both open (stop orders) and sl the same candle - there are those happening when the news are released - I always close the trade as loss even though in many situations the order wouldn't really have hit the sl. The same thing with the very rare events when sl and tp are hit on the same candle.

By opens under/over trade open price I referred to the gaps, ie. those situations where the next 1 min candle open is somewhere higher/lower than the past candle high/low and for example higher than my buy stop (or short sl which is basically buy stop). On those situations I again take the most negative route and execute the stop per candle open, not per the specified stop price so there is some slippage.

I think that should be accurate (or a bit more negative than real world), if anyone can point errors in above, I would be really happy to hear because I want realistic results - I think adding 1 pip to all spreads should cover the slippage very well (that this may be a bit too harsh).
  • #347
  • Sep 3, 2009 1:42pm Sep 3, 2009 1:42pm
Sounds good to me, I remember writing the same type of code.
Once you start to think about it, the whole sequence is rather subtle.
I used to compare my results with MT4 set to 'open prices only' with the same data for debugging.
The breaking of a wave cannot explain the whole sea.
  • #348
  • Edited 4:07pm Sep 3, 2009 3:54pm | Edited 4:07pm
Here is one of the latest pics, now I have a system that stores all found and not working strategies (I decided to call them that, clearer than "algo" which could mean any algorithm) to the database so I don't have to run them all over every time I start the system. Next I'm going to study if I can use the database for correlation checking between strategies by comparing trades they have made.

The risk is now adjusted based on the number of strategies so exponential curve and drawdown is quite ok too. Also the in-sample period is now different color (a bit darker) so the difference between in-sample and out-sample can be seen more clearly now.

I'm quite happy how this is progressing.
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot-AlphaFactory-43.png Size: 136 KB
  • #349
  • Sep 4, 2009 7:00pm Sep 4, 2009 7:00pm
Quoting Craig
I while ago before I had my ATS I tried implementing the SOBI strategy using NinjaTrader on IB, I never really reached any firm conclusions.
I've often thought the book had good potential for AI applications, since it has a much smaller state space than price.

Time to put on the thinking cap.
I looked a bit further into this using data from the order books for a couple of futures. I found that a SOBI approach could be useful for things like estimating the "true price", however it does not seem to be useful as a direct trading signal (based on my admittedly brief research). The distributions looks very stable. In fact, it continues to amaze me how efficient the futures market seems to be on a microscopic scale.

Can you comment on why you did not reach any conclusions? (I can understand if you won't).

The attached paper (in german) describes in a bit more details one of the SOBI strategies that participated in the Penn-Lehman competition.
Attached File(s)
File Type: pdf 02-Steinert.pdf   265 KB | 763 downloads
  • #350
  • Sep 4, 2009 7:15pm Sep 4, 2009 7:15pm
I just wasn't very systematic with my approach, it was my first attempt at a NinjaTrader strategy so much of the time was spent learning the platform (and arguing with NinjaTrader support over bugs). By the time I had it going I had realized that NinjaTrader wasn't going to cut it in terms of a platform and I moved on to starting to construct my current approach, so I guess the book stuff just got lost in the churn.

The basic strategy was SOBI, I ran it against Australian SPI futures index for a couple of weeks, I seem to remember it pretty much bombed, but this was all mixed up in the above.

I really need to take a more structured approach with my current platform.
The breaking of a wave cannot explain the whole sea.
  • #351
  • Sep 5, 2009 6:38am Sep 5, 2009 6:38am
Quoting BlackMage
I looked a bit further into this using data from the order books for a couple of futures. I found that a SOBI approach could be useful for things like estimating the "true price"
This is not directly related to this method but the book in general.

I would claim that book is a bit problematic for precise things because there are many orders that are baits that suddenly disappear when price starts to move towards them - I assume it's some of the big boys who want to fake s/r levels and are willing to take some loss from that. There are certain things that are harder to fake like massive s/r levels but overall I would only use book for a small amount of strategies.

I player with the book a bit during last christmas holidays (I think it was some fx future) when the volume was really low, I put some orders to book above/under the current price and removed them in under second, did this again and watched how the other prices reacted to my orders - there were lots of orders that disappeared the same millisecond as I put my order and reappeared when I removed them. When I repeated the same sequence, exact the same thing happened (no executions were done).

This doesn't mean that you can't use the book data but it's just important to realize that many orders aren't really there in the true sense of someone wanting to buy/sell at some price - there are lots of players who are manipulating the book.
  • #352
  • Sep 5, 2009 3:46pm Sep 5, 2009 3:46pm
All valid points, keep in mind however that the 'price' also plays similar tricks in terms of fake-outs etc.
The breaking of a wave cannot explain the whole sea.
  • #353
  • Sep 5, 2009 4:14pm Sep 5, 2009 4:14pm
Quoting mikkom
it's just important to realize that many orders aren't really there in the true sense of someone wanting to buy/sell at some price - there are lots of players who are manipulating the book.
Very true - the actions in the order book are mostly created by algos who tries to game other algos (or a flesh trader once in a while). But that only makes things so much more interesting.

However, the evolution of the order book over time is the real interesting part (and has almost become a research field by itself).
  • #354
  • Sep 5, 2009 6:58pm Sep 5, 2009 6:58pm
Tim Ord has written some meaningful literature on volume :http://www.tradingmarkets.com/.site/...les/-76494.cfm http://www.tradingmarkets.com/.site/...utor=Tim%20Ord
  • #355
  • Edited 3:00pm Sep 7, 2009 1:37pm | Edited 3:00pm
To change the subject a little, does anybody have any experience with http://wl4.wealth-lab.com/cgi-bin/WealthLab.DLL/getpage?page=Top25.htm
There are strategies listed which have done very well according to the page, yet I assume this is paper trading and the performance does not translate into real life or every WealthLab user would be rich. Anybody know anything more?
The breaking of a wave cannot explain the whole sea.
  • #356
  • Edited 3:31pm Sep 7, 2009 2:43pm | Edited 3:31pm
Quoting Craig
To change the subject a little, does anybody have any experience with http://wl4.wealth-lab.com/cgi-bin/We...age=Top25.htm?
There are strategies listed which have done very well according to the page, yet I assume this is paper trading and the performance does not translate into real life or every WealthLab user would be rich. Anybody know anything more?
You have extra '?' at the end of the line so the link doesn't work.. Anyway, what I find quite funny is that many of these are long only systems and if you look at the stock charts, it's quite clear why those did so well.. I would even claim that any long only system is curvefitted.

I think there is a good possibility that there is something on these algos that can be exploited but it's buried under piles of "optimized" (= curvefitted) algos. Maybe a bit different thing but I have tested probably hundreds or thousands of ea's available at different sites and at least from them I have never found one EA that I would trade live.

The good thing is that the code is there so it's quite easy to test the scripts on multiple markets and see if there is anything there - I would guess that there are couple of things there that are worth studying but to find them you would have to go through hundreds of strategies. I did used to go through sites like that and test every possible strategy but nowadays I prefer books about strategies.

This has nothing to do with the above but here is one of the latest runs with some hundreds of strategies from my platform - nothing special added, just a bunch of different types of algo parts and working order size scaling for multiple strats.

Now the real question is how I can go live with this, I would really like to use a market maker that offers 0.01 lots and increments but the problem is that I don't think there are any that can be trusted - to run hundreds of strategies simultaneously I really have to use a very little risk per trade. The other alternative is to limit the strategy portfolio so that's probably the way to go.

I could also try to find a prop firm and trade with them or start a fund - I don't really think that this is a realistic way at the moment but if anyone has comments about either of these then please post them.
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot-AlphaFactory-48.png Size: 21 KB
  • #357
  • Edited 3:17pm Sep 10, 2009 9:27am | Edited 3:17pm
Here is a pic from today - not much difference in what it looks like but very different inside.

This version now has a simple correlation filter, basically the curve should get smoother the longer I run this (this one is about 20 minutes old). As I assumed, now I can get much smoother curves with fewer strategies.

It's interesting to note how similar this curve looks to the previous one (the risk adjustment is different, this has lower compounded risk if you wonder why some numbers are so different) even though all runs are totally separate and random - this is because the base algos are a bit similar until the system finds something different - and now with correlation algo, this will happen more likely.

I think I'll test risk randomization too with the same framework as with correlations, now I always use the same risk for every algo.

edit: added a second picture of a little tweaked selection algorithm, this is pretty good. I think the next step will be some kind of trading implementation, I think first one I'll made will be an algo to EA converter.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: Screenshot-AlphaFactory-50.png Size: 20 KB
Click to Enlarge Name: Screenshot-AlphaFactory-52.png Size: 18 KB
  • #358
  • Sep 10, 2009 4:01pm Sep 10, 2009 4:01pm
A bit lengthy but nevertheless rather interesting seminar video:

http://www.thalesians.com/finance/in...n_Times_Faster

Basically its about using SimuLink for rapid trading model development.
  • #359
  • Edited 10:47pm Sep 10, 2009 10:45pm | Edited 10:47pm
Quoting mikkom
Here is a pic from today - not much difference in what it looks like but very different inside.

This version now has a simple correlation filter, basically the curve should get smoother the longer I run this (this one is about 20 minutes old). As I assumed, now I can get much smoother curves with fewer strategies.

It's interesting to note how similar this curve looks to the previous one (the risk adjustment is different, this has lower compounded risk if you wonder why some numbers are so different) even though all runs are totally separate...
Hi Mikkom,

Having worked as a financial journalist, an entrepreneur and a business consultant, I firmly believe that the whole forex market is engineered and that there is a definite pattern to it. If by watching statistical charts ( I am not a tech guy) I am able to catch the pattern why can't a tech guy do it? As a businessman my primary focus is on risk control (fear control) which I have succeeded by employing +50-year-old Hedging techniques (buying and selling of same instruments) used by Asian traders for all markets. Europeans and US traders view hedging with contempt.

You must have some patience to watch the charts for one year and log the movements for each half hour, each day, each week, each month, each quarter, each year, each two years and each three years and the patterns will be crystal clear to you.

PJ
Attached Image (click to enlarge)
Click to Enlarge Name: 8 majors doing well.gif Size: 54 KB
  • #360
  • Edited 11:43pm Sep 10, 2009 11:41pm | Edited 11:43pm
The market makers have built buy positions which they will clear today to book profit at higher levels 1.4625 - 1.4660 and once they have liquidated their buy orders they will start short selling. Monday will be a SS (short selling) day.



PJ
3 traders viewing now