Systematic trading | Page 5 | 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

  • #81
  • May 27, 2009 12:45am May 27, 2009 12:45am
Quoting Craig
Just to clarify, the chart is saying that the daily high most often occurs at hour 0 (gmt + 1), so hour 23 gmt (11 o'clock in the evening London time). Isn't this about Asian session lunchtime?
http://www.forexfactory.com/showthread.php?t=168289
The future depends on what we do in the present.
  • #82
  • May 27, 2009 1:42am May 27, 2009 1:42am
Quoting compro99
http://www.forexfactory.com/showthread.php?t=168289
So 0 hours @ GMT + 1 is the Australian market open.
The breaking of a wave cannot explain the whole sea.
  • #83
  • May 27, 2009 1:58am May 27, 2009 1:58am
Quoting mikkom
Here is another interesting chart. This one is from "Bollinger on Bollinger Bands" (good book, worth buying if you don't have it). This is the reason why breakout methods work on market data but wouldn't work on randomly generated data

(This is stock market data - I assume forex data would be even more non normally distributed)
Run a google on The Normality Assumption and Fractal Nature of Asset Prices and you'll find (first link) some stuff that might interest you. All the data on this small project are from fx data.
Measure twice, cut once
  • #84
  • May 27, 2009 2:12am May 27, 2009 2:12am
Quoting Craig
So 0 hours @ GMT + 1 is the Australian market open.
Just something to note, the hour 0 at the chart I did means that high is hit during that hour (I'm not sure if the definition of hour was clear, it could mean that the high was hit nearest of hour X)

I personally think this has not so much to do with any specific session opens, it has much more to do with daily bar open as massive amounts of daily traders [and especially algos] are watching gmt charts and bar opens. IMHO it's the same thing with weekly charts and weekly high/low distribution.

I haven't done any study to how price reacts to daily open at later time of the day but look at the charts and see visually what it looks like - this is something I'm planning to research but I'm not just sure how to quantify the effect. One possibility would be to do a chart from distribution on how far price usually forms a fractal that bounces opposite direction from daily open. I think that would be a very interesting chart that would reveal something.

By the way I was planning to do a matrix that would tell when the opposite ends of the candles are usually met meaning a matrix that would tell what the distribution of highs is if the low is at hour 1-23, the problem was that my mysql froze totally so I had to kill the query after couple of hours. I'll have to try to optimize it a bit..
  • #85
  • May 27, 2009 2:15am May 27, 2009 2:15am
Quoting aediaz1
Run a google on The Normality Assumption and Fractal Nature of Asset Prices and you'll find (first link) some stuff that might interest you. All the data on this small project are from fx data.
Thanks, I'll check it. I'm also thinking of doing a majors distribution chart myself. I want specifically see what really is the distribution between majors, ie. which ones have the highest kurtosis.
  • #86
  • May 27, 2009 2:27am May 27, 2009 2:27am
Here is some more interesting research - when we go into fractal nature of markets this is something everyone should take into account

http://www.forexfactory.com/showpost...6&postcount=20
  • #87
  • May 28, 2009 10:37am May 28, 2009 10:37am
Interesting thread. I am working on the following and would like to know if anyone has any trading ideas.

I am building a program that collects tick data (currently for the Majors). From this tick data I build up up the bars for 1min, 5min, 15min, 1hr, 3hr, 1day. For each time frame I calculate the indicators like EMA, SMA, Stochastic,HAS etc

I store all this information in a database. With a switch I can either trade in real time actually submitting orders or run it against the database for simulation. This simulation is quite accurate as it uses tick by tick data taking spread into account (sampled)

I have the ability the build strategies(multiple) and test them against the four majors. My strategies tend to be structured like this example:

Buy Condition (for example

if (price>3hr.high (use different time frames)
&& price>15min.high
&& 3hr.ema10IsUp
&& 1min.ema10IsUp
&& time>0800
&& time<1200
)
then
BUY
end
stopLossCondition
if loss=20(pips)
SELL
end if;
TakeProfitcondition
if profilt 40 pips
then
SELL
end if;

Ofcourse it can be as complex as you would want it to be. I then run the strategies over the four majors and run a report for the result.

Have to say results are disappointing and usually strategy might make profit on one currency and would loose on the other.

Or winning Strategy with little tweek becomes loosing strategy. You'd expect there would be some "winning statistical" rules that apply to all currencies.

Anyway any ideas are welcome.
  • #88
  • May 28, 2009 2:18pm May 28, 2009 2:18pm
Quoting spekkiefx
Have to say results are disappointing and usually strategy might make profit on one currency and would loose on the other.
You have lots of testing ahead but believe me - there are lots of different strategies that are profitable on all majors.
  • #89
  • May 28, 2009 4:36pm May 28, 2009 4:36pm
I think what you need now is an optimiser module. Something that will do automatically and much faster what you now do manually i.e. try many different combinations of strategy rules with different parameters, instead of you having to write the systems yourself individually. Since you've come so far as having the database/feeds module and simulation module in place, you shouldn't have a problem creating the optimiser, at least a simple one.

Quoting spekkiefx
Anyway any ideas are welcome.
  • #90
  • Edited 5:29pm May 28, 2009 5:17pm | Edited 5:29pm
Quoting challenger78
I think what you need now is an optimiser module. Something that will do automatically and much faster what you now do manually i.e. try many different combinations of strategy rules with different parameters, instead of you having to write the systems yourself individually. Since you've come so far as having the database/feeds module and simulation module in place, you shouldn't have a problem creating the optimiser, at least a simple one.
Thanks Mikkon/challenger for the advice!

I have thought about that but currently my thoughts have been to use the charts to spot tendencies and then attempt to put them into rules. My thinking was that optimization shouldn't really matter to pick out these tendencies. Especially as the market changes constantly. (not sure if I still believe this as it is clear to me that a maximum stop loss differs per currency pair considerably)

One interesting conclusion I have come to is by describing price patterns I can achieve the same or better results without using any SMA/EMA and stochastic, indicator stuff etc. This supports the idea that these are lagging indicators and have little use. Therefor not much point to optimize these.

The good thing of my approach is that I have very accurate results taking spreads into account. The bad thing is that currently I have limited data.

Also interesting that for non-major currencies it is really difficult to have any successful system as they almost always tend to loose.
  • #91
  • May 29, 2009 3:42am May 29, 2009 3:42am
Quoting spekkiefx
Have to say results are disappointing and usually strategy might make profit on one currency and would loose on the other.
Why should a strategy necessarily apply on all majors? Spreads, liquidity, volatility and so on differs a lot.
  • #92
  • May 29, 2009 4:23am May 29, 2009 4:23am
Quoting BlackMage
Why should a strategy necessarily apply on all majors? Spreads, liquidity, volatility and so on differs a lot.
My view is that there must be certain patterns that "people" are looking for. This is what TA/Gann/Pivot etc are all about. Basically patterns that will give you a statistical edge, I 'd expect them to be the same across the board as they are the result of "herd" behaviour of humans.
  • #93
  • May 29, 2009 5:13am May 29, 2009 5:13am
Quoting spekkiefx
My view is that there must be certain patterns that "people" are looking for. This is what TA/Gann/Pivot etc are all about. Basically patterns that will give you a statistical edge, I 'd expect them to be the same across the board as they are the result of "herd" behaviour of humans.
Yes, but what if this edge can overcome market frictions in one market but not in another.
  • #94
  • May 29, 2009 6:35am May 29, 2009 6:35am
Quoting BlackMage
Yes, but what if this edge can overcome market frictions in one market but not in another.
For me it's just a part of my verification process. If a (simple) strategy works well on all majors, it almost definitely is not curve fitted. If it works just on one pair it might be curve fitted. Someone else might do a verification totally otherwise but that's how I decided to do it years ago. There are also some big plusses with this approach, for example if some pairs are not correlated, the combined equity curve for all currencies is much smoother than it would be with one currency.

There definitely are some [especially intraday] strategies that work differently on different currencies because liquidity is spread differently for yen pairs than chf for example - I don't trade intraday so that's not a concern for me.
  • #95
  • May 29, 2009 8:03am May 29, 2009 8:03am
Quoting mikkom
There definitely are some [especially intraday] strategies that work differently on different currencies because liquidity is spread differently for yen pairs than chf for example - I don't trade intraday so that's not a concern for me.
Funny you should say this. This is exactly what I am working on. I am implementing the ability to have strategies span for more than 1 day. Currently my program only allows for intraday. So once a strategy is outside the days "noise" it should hopefully show better results.

Seeing is believing(-:
  • #96
  • Jun 3, 2009 8:59am Jun 3, 2009 8:59am
I have included the ability of my program to span over multiple days. This has improved results. Had another go at building a system and managed to come up with a system that managed positive results for 7 out of 8 different currency pairs(never achieved that before). Only 3 weeks of data though so be interesting to see what happens.
  • #97
  • Jun 4, 2009 6:19am Jun 4, 2009 6:19am
How many trades on each pair in the 3 weeks?

Quoting spekkiefx
I have included the ability of my program to span over multiple days. This has improved results. Had another go at building a system and managed to come up with a system that managed positive results for 7 out of 8 different currency pairs(never achieved that before). Only 3 weeks of data though so be interesting to see what happens.
  • #98
  • Jun 4, 2009 9:29am Jun 4, 2009 9:29am
6 out of 7 are positive
Currency #positions
EUR/USD 9
GBP/USD 9
USD/CHF 14
EUR/GBP 13 Negative result
USD/CAD 12
AUD/USD 11
NZD/USD 7

Statistically not that many but for different currencies though!!! Does this strategy has the edge? It is getting interesting now! Especially as I use tick data(synchronised) and results "should be" accurate.
  • #99
  • Jun 4, 2009 9:59am Jun 4, 2009 9:59am
Quoting compro99
May i know yr entry/exit rules ? Holding period... ?
Please, no specific system rules here. This thread is about systematic trading not any specific systems. If you want to discuss about specific systems start a new thread on trading systems forum.

I would say that the sample size that spekkie gave is so small that the results unfortunately tell nothing.

Thanks.
  • #100
  • Jun 5, 2009 3:49am Jun 5, 2009 3:49am
Quoting mikkom
I would say that the sample size that spekkie gave is so small that the results unfortunately tell nothing.

Thanks.
I agree it is not a large sample. But it would be equivalent to 7(pairs)*3 weeks =5 months (1.4 GB of tick data) and as discussed no Curve Fitting. Also spreads are taken into account! You look at periods over 10 years what is the smallest time frame you use? How accurate are those results compared to real trading?

I am sure the performance would not be as current figures or I would be ordering a yacht. However there is some room to go down and still be a good system, anyway time will tell.

Any thoughts of playing the 7 main currencies at the same time with one strategy?
0 traders viewing now