data mining in tradestation?

[darthtrader 10]

i've really been trying to get into learning easylanguage the past few weeks and had a post all typed up yesterday in the candlestick thread after seeing common candle patterns as reserved words in ez. The post was pretty unclear though so i deleted it. Surfing around then I found the data mining webinar on the tradestation site, which was basically i guess what i was going to ask if tradestation could do.

Has anyone ever done any type of data mining with tradestation? I was thinking something along the lines of mining for candle patterns around a setup.

I also think it would be interesting to mine for an optimal stop/profit target on a setup. I mean how do you know that a 22 or a 17 point stop is not actually better than a 20 point stop?

Anyone use this stuff or find that its a huge waste of time?

Also, anyone ever do anything with the probability maps in TS? Seems like that should be usefull somehow with market profile.

[waveslider 10]

Hi Darth,

 

Data mining is generally a waste of time. It is the same as curve fitting, and for most circumstances un-useable. The reason is that that market is dynamic and irrational, just like the human mind.

 

I have been around easy language for years, almost everything I have developed I have tossed. The reason most tools are not helpful is because nothing competes with the human eye.

 

If you are trading stocks it can be helpful to program in logic to call attention to certain conditions with easylanguage. My opinion is that easylanguage is useful to test very general ideas, but at all times runs the risk of distraction.

 

You can be sure that just about every worthwhile idea you could come up with is begin exploited - especially candles.

 

I think there are very few tools that you need - everything is really just an extraction of price and volume. Tools like market profile are a good way of organizing information, but the same information is available there to the naked eye.

 

The probability map seems to be garbage also.

 

I think I can speak for a large number of people who can look back and say they have spent way too much time programming, time that could better be used refining and practicing one approach.

 

hope that helps!

 

ws

[torero 10]

I'm torn on siding with or against ws' comments. It's true it takes alot of time to come up with something useful enough to even consider trading it real time. Even then, these are not suitable for daytrading (swing or higher). Having said that I've found a few systems through data mining but took me a 2 years drudging at it. So, if you have time and 2 years, then learning this side of trading is very valuable. These data mining finds have to be done with extreme care (ie. very skeptical) or it'll be your $$$ with a wrong assumptions or operating with full knowledge of the matter involved.

[darthtrader 10]

Interesting responses to think about guys.

The time involved with this isn't an issue, i took alot of programming classes in college and i'm finding it so much fun to actually be able to program something that actually does something.

I soppose it is something I will not exactly concentrate on but just explore when i have extra time.

Figured the probability maps were useless just being i've never found anything that uses them, shame because they do look cool.

 

 

thanks guys.

[waveslider 10]

Hey Darth, have fun - enjoy! It's great to have a program present the market to you the way you want to see it. My only point was that you shouldn't expect to find the key to the markets through data mining.

I agree 100% with Torero, there are some valid methods to be found using the daily time frame. I manage a fund that I trade only on daily signals, the system I developed to trade the fund is 90% mechanical.

Also, as he mentioned, you have to be very critical of any results you may come up with. Testing them on virgin data, un-correlated markets, etc.

 

ws

[torero 10]

Looking back, my first mistake when I first got involved with program trading was I got so excited seeing all these $$$ after running a few tests I put it on real trading right away. Took a week to realize what I saw and what I thought I saw was completely unrealistic. Lesson learned: if it looks good and easy, double-check, triple-check and test it from all angles and read the forums on program trading as much as you can before taking it for a real spin. There are tons of traps imaginable and unimaginable.

[darthtrader 10]

good stuff.

i guess in some sense though i'm less interested in finding a mechanical system and more so interesting in mining for information around a discretionary setup.

Thats where i thought the candle pattern stuff might be interesting, a full setup you would normally trade discretionary(without looking at candles) but then maybe mine to see if if there are any candle patterns that occured better than chance. Not that i would then make that mechanical but just something to look for. Or if volatility is X, mine to see what the average move was or the average loss was from the setup, more things like that.

 

did you guys use any kind of volatility measure in your systems? I've been trying to think what the best measure of volatility would be, implied volatility on the index options?

[torero 10]

I used a few things to measure volatility: price channels, ATR, Bollinger bands, and Keltner. They are useful in some ways but not in others.

 

About candlesticks, one thing I've done wrong with them was identifying setups based on a single or 2 candlesticks. These i found out led to alot of false signals. In mining, it's better to group the bars at least 5 or more and describe them in more details before the false signals are taken out. One or two bars cannot say much when 50-60 bars have already been voted, even if they are tops or bottoms.

[BlowFish 10]

I can't help feeling that it could be useful. Knowing that the high/low of the day is put in during the first 15 minutes of trading 75% of the time (figures made up on the spot) should be exploitable. Knowing that price trades withing 5 ticks of yesterdays floor pivot 70% of the time likewise. Knowing that if price travels yy points above last weeks high that it is likely to go on and close (daily) above this value xx% of the time etc. etc. etc.

 

I guess what I am saying is things should be statisticaly significant (i.e. a good high number personally I'd like to see 70% or even 80%+). And that the 'thing' you are measuring should make some sense based on market structure. I guess you have to be pretty careful that you are uncovering some fundamental structure rather than curve fitting some random numbers of course.

 

I have only ever tried a data mining exercise once as the tools are a little unwieldy (I used Excel). If there where simple tools I might be inclined to try more.

 

Cheers.

[torero 10]

I am not saying you shouldn't look into it. I encourage it and find the answers for yourself. Just my limited observation on this area that steered me to other approaches.

[darthtrader 10]

blowfish, thats absolutely the type of stuff i'm interested in. I would think it would be interesting for extra confirmation as opposed to trying to use things like that outright in and of itself.

 

you guys ever try trade ideas? I believe thats what Brett Steenbarger uses to do stuff kind of like this, i'm not sure how deep that gets though as far as on more microstructure type stuff.

http://www.trade-ideas.com/DOSA/

 

i would also think this type of stuff would work better if you had a decent measure of volatility and some rough measure if the market is trending or range bound. if you had something like that you could at least make sure your in the ball park for current conditions vs past conditions.

 

candlesticks were really just an example i threw out. i agree with looking at 5 bars or what have you. i soppose the problem there would be how fuzzy the patterns would be and would probly be quite hard to program anything usefull in that area.

[rvlv 10]

Thank you Blowfish.

I have recently started with Tradestation 2000i . Still confused about what indicators to use,what guidelines to follow.

 

Your suggestion of pivots is good and I begin to explore that.

 

As related to formulae, I found two things really interesting to me which I like to share with experienced people over here. I like to swing trade.

1. Jim berg formula

ref link

http://www.traders.com/Documentation/FEEDbk_docs/ Archive/022005/TradersTips/TradersTips.html

 

2.support-resistance formula

Support and resistance levels using RSI.

BY Graham Kavanagh May 2003 Ref Amibroker library

I dont have this written in easy language yet, trying still.

The concept is as I see

 

Go for the easiest AND simple formula there in, Wait for the mkt to come to support levels AND Buy, wait for the mkt to reach resistance level AND Sell

take your bucks.

The point I keep wondering is- do I chase the market? or do I wait till market comes to my choice levels of support and resistance?

 

I am a newbie and looking forward to seeing useful,dependable indicators and formule that can be applied in tradestation. My friends tell me go for JMA ,ASC trend etc.

Please advise ways to shorten learning curve in TS application-

advance thanks for your time

rvlv

[BlowFish 10]

A couple of random thoughts.

 

Someone once told me that in the ES traded at the floor pivot about 70% of the time (within a tick or two on a within the day basis). I got 68% when I put the numbers in Excel. Now that same person also used the pivot as a crude directional indicator above the pivot bullish below bearish. Fair enough but this clearly is not behaviour that could be expected from the test. (We would need to measure something like where the close would be compared to the open compared to the pivot)

 

Seems to me that a strategy could be built to trade a return to the pivot based on the fact that 68% of the time you would get there.

 

Opening range break outs are a strategy that hopes to capitalise on a certain market behaviour (un-confirmed by me) and that is that the high or low is 'often' put in in the first xx minutes of trading.

 

Still no real thoughts on tools - Excel will do the job to a degree but it would be nice to have something simple and fairly non programmatic to use.

[cajam 10]

Not sure if this thread is still active but in case someone wanders in. I have done quite a bit of data mining with easylanguage. Developing mechanical systems based on EOD data. From my experiences I found a few tendencies that can be turned into decent mechanical systems (that meet my own criteria anyway!). The key I found was keeping it as simple as possible. For example the best I found was not using any indicators and just looking at the HLOC relative to each other over 2 to 5 days (eg. Gaps). Other interesting tendencies come from volatility analysis (range expansion or contraction). IMO you need results that give you >50-60 occurrences per year and that work without too much variance for 10 years + or even 20 years (Gaps).

 

I also agree there are many traps but a little bit of research can overcome this (check the link below!).

 

There are some successul traders using this approach at http://www.thechartist.com.au/forum/ubbthreads.php and a wealth of info.

 

Cam