Systematic trading | Page 46 | 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

  • #901
  • Jun 2, 2010 2:09am Jun 2, 2010 2:09am
Quoting mikkom
I never turn my strategies on/off, however strategies themselves might have some rules that tell when they should trade. I think it must be strategys decision to choose when to trade, not some external systems - basically every one of the external system rules should be possible to add to original system if they make results better.

I don't categorize my systems, [selective] correlation is what really matters, not what kind of system it is or how I have categorized it.
I totally agree with this strategy!!!
Could you please specify what platform you use for auto trading?
Entry is a science, exit is an art.
  • #902
  • Jun 5, 2010 8:40pm Jun 5, 2010 8:40pm
Nice thread guys.

Can anyone give me some advice on using an API?

Are there open source terminals which I could use for it or would I need to learn to program my own interface? If so what would be some good books/websites to read on it.

I have barely any programming knowledge but want to learn.

And for you guys that are always on the look for more historical data, http://eareview.net/tick-data

Free historical tick data with instructions on how to download and parse it efficiently.

Regards Seb.
The greatest force on earth is compound interest.
  • #903
  • Jun 6, 2010 3:42am Jun 6, 2010 3:42am
Quoting ForexQuant
Their individual equity curve is quite choppy but the good news is that their correlation among each others is from -10% to 45%.
Forexquant, how do you calculate the correlation. Is it based on the equity curve created through trade by trade or at time (daily) basis?
  • #904
  • Jun 6, 2010 8:41am Jun 6, 2010 8:41am
Quoting sebastionay
Nice thread guys.

Can anyone give me some advice on using an API?

Are there open source terminals which I could use for it or would I need to learn to program my own interface? If so what would be some good books/websites to read on it.

I have barely any programming knowledge but want to learn.

And for you guys that are always on the look for more historical data, http://eareview.net/tick-data

Free historical tick data with instructions on how to download and parse it efficiently.

Regards Seb.

Thanks for the data info, havn't heard of that site before but it has some helpful scripts.
  • #905
  • Jun 6, 2010 4:59pm Jun 6, 2010 4:59pm
Quoting poekmeister
Forexquant, how do you calculate the correlation. Is it based on the equity curve created through trade by trade or at time (daily) basis?
both.
  • #906
  • Jun 6, 2010 9:30pm Jun 6, 2010 9:30pm
Quoting sebastionay
Nice thread guys.

Can anyone give me some advice on using an API?

Are there open source terminals which I could use for it or would I need to learn to program my own interface? If so what would be some good books/websites to read on it.

I have barely any programming knowledge but want to learn.

And for you guys that are always on the look for more historical data, http://eareview.net/tick-data

Free historical tick data with instructions on how to download and parse it efficiently.

Regards Seb.
I'm roughly in the same boat as you are, in that I'm building up my own little system from scratch, and trying to reuse as many open-source chunks as possible to make it easier.

Regarding the dukascopy tick-data, I'd recommend to not convert them to CSV files, as it's both much larger on disk and slower to use.
Stick to the download php script, although I'd suggest unzipping the files you download to make reading them faster. Or you could just leave them alone for now and unzip them on first read.

I bumped into this page http://ta-lib.org/hdr_lnk.html which lists a number of related open-source projects, but I haven't looked at most of them yet.
There should be a number of attempts at writing a terminal in there, although the level of completeness is uncertain.

As far as APIs, I'm guessing most of the folks with a working ATS here are using FIX.
There are open source implementations of it, like http://www.quickfixengine.org/ .

That doesn't cover the account and order management stuff though. I'm still not sure if I'm writing my own thing for that, or if there's some open-source pearl lying around doing exactly what I need already.
How did this get here oh god I am not good with computer.
  • #907
  • Jun 6, 2010 9:39pm Jun 6, 2010 9:39pm
There are some links to open source platforms on the end of the post #1 of this thread.
  • #908
  • Jun 7, 2010 12:31am Jun 7, 2010 12:31am
Nice link krakoukas.
  • #909
  • Jun 7, 2010 6:15pm Jun 7, 2010 6:15pm
What do you guys think of this discussion regarding:

Expectancy Score vs Sharpe Ratio for strategy optimization / selection?

http://unicorn.us.com/trading/expectancy.html
  • #910
  • Jun 8, 2010 12:22am Jun 8, 2010 12:22am
This might be of value (?), just found it - so no idea if it's actually practical in trading domain.

http://ccsl.mae.cornell.edu/eureqa
  • #911
  • Jun 8, 2010 12:38am Jun 8, 2010 12:38am
Quoting FXEZ
What do you guys think of this discussion regarding:

Expectancy Score vs Sharpe Ratio for strategy optimization / selection?

http://unicorn.us.com/trading/expectancy.html
I always prefer formulas that react as fast as possible to fundamental changes in result curve and that take clustering of trades into account in some way.

I personally don't currently use either of the above formulas in my calculations.
  • #912
  • Jun 8, 2010 2:43am Jun 8, 2010 2:43am
Quoting FXEZ
What do you guys think of this discussion regarding:

Expectancy Score vs Sharpe Ratio for strategy optimization / selection?

http://unicorn.us.com/trading/expectancy.html
There are problems with all measures, all represent a trade-off of some description. The most obvious failure of either expectancy or sharpe ratio is the failure to factor in the number of trades. The more something happens the more you would expect the event to continue and the less chance it has of being a statistical artifact. The t-stat addresses this somewhat, but has the same problems as the sharpe ratio in terms of being able to easily construct counter examples which show it's limitations.
The breaking of a wave cannot explain the whole sea.
  • #913
  • Jun 8, 2010 6:33am Jun 8, 2010 6:33am
About the correlation calculations.. To calculate correlation between two strategies, I currently calculate it with pearson based on series of time (minute), I only use points of time where either one of the algorithms has a trade open - each point is pair of [-1, 1] (-1 is 100% of allowed risk short and +1 is 100% allowed risk long, 0 means that no trade is open).

What I have been wondering is should I also include the all-zero points (no trades open on either of the algos) to the series? Are there any stat guys here who can answer this? Is pearson even optimal on this kind of comparison or would some other method be better?
  • #914
  • Jun 8, 2010 6:34am Jun 8, 2010 6:34am
Quoting Craig
There are problems with all measures, all represent a trade-off of some description. The most obvious failure of either expectancy or sharpe ratio is the failure to factor in the number of trades. The more something happens the more you would expect the event to continue and the less chance it has of being a statistical artifact. The t-stat addresses this somewhat, but has the same problems as the sharpe ratio in terms of being able to easily construct counter examples which show it's limitations.
By the definition given in the link, the Expectancy Score = Expectancy x Opportunity, and Opportunity = Non Scratch Trades x 365/(calendar days of history being tested). So from my reading it appears to take the # of trades into account, presumably making it a slightly better metric than plain Expectancy. Were you referring to something else or do I miss your point?
  • #915
  • Jun 8, 2010 7:52am Jun 8, 2010 7:52am
Quoting mikkom
About the correlation calculations.. To calculate correlation between two strategies, I currently calculate it with pearson based on series of time (minute), I only use points of time where either one of the algorithms has a trade open - each point is pair of [-1, 1] (-1 is 100% of allowed risk short and +1 is 100% allowed risk long,...
Personally I do not calculate the correlation for the direction of trade, I just measure the correlation for the P&L between two strategies. Anyway, I am not good enough to answer your question.
  • #916
  • Jun 8, 2010 8:03am Jun 8, 2010 8:03am
Quoting Craig
There are problems with all measures, all represent a trade-off of some description. The most obvious failure of either expectancy or sharpe ratio is the failure to factor in the number of trades. The more something happens the more you would expect the event to continue and the less chance it has of being a statistical artifact. The t-stat addresses this somewhat, but has the same problems as the sharpe ratio in terms of being able to easily construct counter examples which show it's limitations.
I suggest to use skewness beside all these performance parameter. If your result is negative skew then you may easily overestimate your performance and subject to the black swan risk.
  • #917
  • Jun 8, 2010 10:35am Jun 8, 2010 10:35am
I was just wandering what others are doing concerning time used in the ats.

Anyone using gmt? Or does anyone adjust it to BST (GMT+1 -British Summer Time)for the London session (And adjust it for the US session)? I am not sure what is best any opinions?
  • #918
  • Jun 8, 2010 11:30am Jun 8, 2010 11:30am
Quoting ForexQuant
Personally I do not calculate the correlation for the direction of trade, I just measure the correlation for the P&L between two strategies. Anyway, I am not good enough to answer your question.
What kind of series are you using? I have been thinking if my series are too fine-grained and if hourly or daily series would be better.

p/l correlation might be something I might try too, but I'm not sure if I want strategies with negative correlation to my overall curve
  • #919
  • Jun 8, 2010 11:58am Jun 8, 2010 11:58am
Quoting FXEZ
By the definition given in the link, the Expectancy Score = Expectancy x Opportunity, and Opportunity = Non Scratch Trades x 365/(calendar days of history being tested). So from my reading it appears to take the # of trades into account, presumably making it a slightly better metric than plain Expectancy. Were you referring to something else or do I miss your point?
No, I missed that, I just assumed they were using the standard definition of expectancy. Though I'm not sure what logic is behind the 'Opportunity' measure is since they only seem take winners into account(?). So if you test over 365 calendar days all you are effectively doing is multiplying expectancy by the number of winners (sans the largest winner). So it would seem that over longer periods the number of winners would dominate the 'Expectancy Score', I'm not sure that makes any sense, but I could be missing something again...
The breaking of a wave cannot explain the whole sea.
  • #920
  • Edited 12:30pm Jun 8, 2010 12:00pm | Edited 12:30pm
Quoting spekkiefx
I was just wandering what others are doing concerning time used in the ats.

Anyone using gmt? Or does anyone adjust it to BST (GMT+1 -British Summer Time)for the London session (And adjust it for the US session)? I am not sure what is best any opinions?
All my dates and times are stored with a timezone, the ATS then converts everything into EST. This is done using the boost date/time classes.
Edit: If you want to define a trading session which starts in London and ends in NY you will face the problem of the DST changes in the different timezones, this is why I ended up going down this path.
The breaking of a wave cannot explain the whole sea.
1 trader viewing now