Comments on Max Dama on Automated Trading: Decision Trees

Thanks Chris.

2010-04-22T08:38:23.435-07:00

Thanks Chris.

Hi Max, Great blog! Excellent source of material...

2010-04-22T08:31:34.069-07:00

Hi Max,

Great blog! Excellent source of material to browse though and thanks for sharing!

CS

Ken, Just an update: I'll post the next note by ...

2009-05-18T06:47:00.000-07:00

Ken,

Just an update: I'll post the next note by Thursday.

Regards,
Max

Max, check out what the R project has proposed for...

2009-05-16T23:02:00.000-07:00

Max, check out what the R project has proposed for the Google Summer of Code:

http://www.r-project.org/soc09/ideas.html#p3

I'm anxious to read your post on the weaknesses of decision trees and your improvements. I use regression trees quite successfully in trading, and I'm curious whether you'll say anything that's new to me.

Josh, My bad... It works as expected / documente...

2009-05-09T10:09:00.000-07:00

Josh,

My bad... It works as expected / documented then ;-) Just not what I had expected ;-(

I'll be looking forward to seeing the acquire and analyze example that you and Jeff are working on.

Again, thanks to all 8^)

Cordially,

-Digital Dude-

"Simple things should be simple. Complex things should be possible." -Alan Kay-

DD, I just tried it myself without issue. R "han...

2009-05-09T01:48:00.000-07:00

DD,

I just tried it myself without issue. R "hangs" because it isn't multi-threaded (i.e. it can only do one thing at a time).

So, it's a bit tricky to gather an analyze data at the same time. That said, Jeff and I are working on an example of how to run a real-time trading strategy using IBrokers.

Max & Josh, Thank you both very much! Its nice t...

2009-05-08T19:39:00.000-07:00

Max & Josh,

Thank you both very much! Its nice to see the collaboration ;-)

I'll be looking for your upcoming R work Max...

I've already seen this example with IBrokers Josh... It hangs for me collecting data forever... I was looking for a more useful example... As for the model building in quantmod... Just hack something together to get a working start... I'm sure you and Jeff will get plenty of feedback ;-) Rattle seems to have hung some stuff together...

Cordially,

-Digital Dude-

“We keep moving forward, opening new doors and doing new things, because we’re curious.” -Walt Disney-

Josh, As I've been building more models I've noti...

2009-05-08T09:52:00.000-07:00

Josh,

As I've been building more models I've noticed that there's not one perfect template for all systems, so I see how that could be an impediment. I think about it often too so I'll get in touch with him soon.

Regards,
Max

You're welcome Max. The only reason it's easy is ...

2009-05-08T09:48:00.000-07:00

You're welcome Max. The only reason it's easy is because of Jeff Ryan's work on the quantmod package.

Please keep me in mind when you do start using R more. I'm more-than-happy to help. :-)

Sadly, the model building/testing functionality of quantmod hasn't progressed much. One big reason for this is because Jeff isn't certain how they should be implemented. Think of them more as "concepts" than firm ideas that simply need to be coded. If you have ideas for how they should be implemented, please contact Jeff.

Thanks Josh. Everything's easy in R. The main re...

2009-05-08T09:38:00.000-07:00

Thanks Josh. Everything's easy in R.

The main reason was what I mentioned to Digital but slipped my mind when I first replied,
"I've been hesitant to switch because this summer I have to use Matlab at a job so I want to stay in that mindset until it starts."

One question: in the quantmod package, are the model building and testing functions completed? I've tried to use them but got an error somewhere and the examples on the website haven't been finished.

Regards,
Max

Max, In that case, it's pretty easy. This should...

2009-05-08T09:26:00.000-07:00

Max,

In that case, it's pretty easy. This should do the trick:

require(quantmod)
url <- "http://www2.standardandpoors.com/servlet/Satellite?pagename=spcom/page/download&sectorid=%20%3E%20%2700%27&itemname=%3E=%20%271%27&dt=07-MAY-2009&indexcode=500"

# Read in CSV from S&P website
x <- read.csv(url)

# Pull the first 10 symbols from Yahoo
symbols <- paste(x[1:10,'Symbol'],collapse=";")
getSymbols(symbols, env=.GlobalEnv)

Josh, It downloads data only for the current comp...

2009-05-08T09:18:00.000-07:00

Josh,

It downloads data only for the current composition.

Max

Max, Do those Matlab functions download all the S...

2009-05-08T09:16:00.000-07:00

Max,

Do those Matlab functions download all the S&P 500 constituents as-of a particular day, or does it download the data for each constituent as the index was defined for each day (i.e. accounting for index changes)?

Digital Dude,

I will post an example of using this code in an actual trading rule in the next few weeks.

Make sure you have the latest version of R and IBrokers, and the TWS needs to be running.

require(IBrokers)
tws <- twsConnect()
fh <- file('ES_depth_0509.dat',open='a')
reqMktDepth(tws, twsFUT("ES","GLOBEX","200905"), file=fh )
close(fh)

Max, I'm still around and looking forward to more...

2009-05-08T08:40:00.000-07:00

Max,

I'm still around and looking forward to more questions and answers and examples from you and your readers ;-)

Josh,

Any chance you would share some or all of your realtime ES book data with IBrokers and R as an example???

Cordially,

-Digital Dude-

“Weed out the bureaucracies that stifle cooperation and collaboration.” -Jack Welch, ex-Chairman, General Electric-

Digital, I was wondering if you were still aroun...

2009-05-07T23:30:00.000-07:00

Digital,

I was wondering if you were still around.

I agree, also I'm teaching a class at Berkeley on quantitative trading which I'll probably need to use R for since I can't require everyone buys Matlab. However, I've been hesitant to switch because this summer I have to use Matlab at a job so I want to stay in that mindset until it starts. I would like to recode the decision tree system in R anyways. In fact, I've started on it.

Regards,
Max

Max, It would be nice to see your examples in a f...

2009-05-07T18:31:00.000-07:00

Max,

It would be nice to see your examples in a free open source solution (FOSS) so your readers can all follow along...

Cordially,

-Digital Dude-

"We have very few inferior people in the world. We have lots of inferior environments. Try to enrich your environment." -Frank Loyd Wright-

Josh, It downloads all the data over a specified ...

2009-05-06T23:04:00.000-07:00

Josh,

It downloads all the data over a specified period for all the symbols, taken from S&P's website I believe, and filters out the ones that have missing data. Here's the file: http://www.mathworks.com/matlabcentral/fileexchange/23569

I have a funded IB account.

Regards,
Max

When you say "download the S&P500" are you referri...

2009-05-06T21:28:00.000-07:00

When you say "download the S&P500" are you referring to the index or all its components?

If it's just the index, take a look at getSymbols() in the quantmod package. If it downloads all the components, I could quickly write a function to do that in R. I would just need to know how to determine all 500 symbols. :)

I don't know how easy it is to use IB's TWS with Matlab. I've had success downloading real-time ES book data with IBrokers in R. If you don't have an IB account, you can test the IBrokers package with the IB demo TWS.

Josh, Thanks, I've come across the random forest ...

2009-05-06T21:23:00.000-07:00

Josh,

Thanks, I've come across the random forest library for R, and I was going to use it for the system I mentioned above. In the end I decided to use Matlab because there's a good function to download the SP500. My next system will likely be in R because of the IBrokers package. I was wondering how easy it would be to just use IB's Java API in either R or Matlab.

Regards,
Max

Max, This is a nice and quick introduction to dec...

2009-05-06T20:22:00.000-07:00

Max,

This is a nice and quick introduction to decision trees. I've used Breiman and Cutler's random forests to test if my ideas are predictive of the next day's returns. It's quite easy to use in R.

I look forward to your next post on this topic.

Best,
Josh