R-Multiple misleading ?

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Turbowagon
Roundtable Fellow
Roundtable Fellow
Posts: 82
Joined: Wed Oct 08, 2003 1:19 pm
Location: Washington, DC

Post by Turbowagon » Wed Jul 18, 2007 9:18 pm

Sluggo wrote:
However it may be wise to remember that "most of these guys" is not the same as "all of these guys". Go visit 600 Route 25A, East Setauket, NY, 11733 USA for a reminder.
They just happen to be hiring,... you interested?

http://tinyurl.com/348puk

Regards,
eD

AFJ Garner
Roundtable Knight
Roundtable Knight
Posts: 2040
Joined: Fri Apr 25, 2003 3:33 pm
Location: London
Contact:

Post by AFJ Garner » Thu Jul 19, 2007 2:31 am

http://www.tower-research.com/Careers.html

There is another one for you. And another exception.

Old European
Roundtable Knight
Roundtable Knight
Posts: 144
Joined: Mon Apr 19, 2004 3:04 pm
Location: Old Europe

Post by Old European » Thu Jul 19, 2007 3:01 am

I think this is a discussion about semantics.

What is simple and what is complicated?

A simple concept like 'risk-adjusted return' may look complicated to many, while a scholar with a degree in statistics and applying for a job with Renaissance Technologies may be surprised to find out that Jim Simons' statistical models are surprisingly simple.

Personally I prefer to make a distinction between what works and what doesn't work.

Look at forex carry trading (borrowing in low yielding currencies like JPY, selling these currencies against high yielding currencies like NZD and depositing the high yielding currencies). It's an incredibly simple strategy (a straightforward exercise in risk management), but it has been extremely profitable over the last few years.

Take another example. Programming an algorithmic high-frequency forex trading strategy requires a lot of hard and dirty work (collecting real-time data and building a database, programming a testing engine in Java, doing extensive testing and developing an API to fully automate trade placement and bookkeeping, ...), but in the end it can be very rewarding.

I agree that simple models (with less degrees of freedom) are less prone to curve-fitting and are therefore preferable. But even more important is to focus on systems that work, to be constantly on the lookout for possible system breakdowns and to work on alternative models.

Cheers,

Old European

AFJ Garner
Roundtable Knight
Roundtable Knight
Posts: 2040
Joined: Fri Apr 25, 2003 3:33 pm
Location: London
Contact:

Post by AFJ Garner » Thu Jul 19, 2007 3:25 am

surprised to find out that Jim Simons' statistical models are surprisingly simple
Much what I was trying to say myself but far better expressed. Perhaps more complex than simple TF and certainly far more complex to execute but perhaps, at heart, not so out of this world.

And, I suspect, the concepts can probably be adequately explained in jargon free language!

Chuck B
Roundtable Knight
Roundtable Knight
Posts: 469
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B » Mon Oct 08, 2007 9:37 am

christianh wrote:AFJ Garner, I agree with you, but still the original question is the validity of a change to to the R-multiple calculation because of its shortcoming. I like the concept of R and have built trading models around it but...

"Wouldn't it be better to change the R-multiple calculation to take not only the initial risk but also the ongoing open risk (whichever is higher) into its calculation?" So the new formula would be:

R = Return / Max(Initial Risk, Max Open Risk in Trade)


IMO, this new formula is "better" than the original one. I would like to hear others comment on it (flaws in the formula,...).


Christian
Christian,

I'm the one who created the R-multiple as a concept many years back. At the time all trading software (such as Trading Recipes) referred to expectation in dollar terms which I found next to worthless. A friend mentioned how the Turtles used volatility at trade entry to measure the progress of a trade and referred to them as "multiples of N" (where N was something like the 10 or 20 day ATR). I jumped onto looking at how trades in my systems looked in terms of their initially defined risk (since my main system used a volatility measure at entry).

I programmed Trading Recipes to spit out into a column the multiple of R throughout a trade along with the closed R. Since all trades were sized as a percentage of the portfolio equity (realize this was almost unheard of stuff in 1993-94 timeframe), I now had a method of equating the returns of a given trade that was independent of the specific market. Prior to that, stuff like Tradestation looked at only "dollars profit" based on the contract size of a specific market, etc (and still do by the way, 15+ years after Tom Basso tried to get Bill Cruz to create a "real" piece of trading software).

So the R-multiple was created to essentially normalize any given market to the whole basket of markets being analyzed. It didn't matter if it was a corn contract with a $15,000 face value or the S&P with a $200,000 face value -- they all spit out the same relative data, in terms of R. Again, it was important to reference to the initial risk, R, since that was what was being used to size the trades in relation to the account equity.

Once I reached that step, I then saw how I could now see deep into "how" a given system generated the equity curve outcome seen only as a continuous, bar-by-bar, curve. It then allowed expectation to be based on R instead of the worthless "dollar" measure. It showed how much return, based on the initial risk in a trade, a given system made along with how much "total open R" the system exit strategy captured. All my initial research looked at total open R on a bar-by-bar basis versus closed R in the trade. I developed many additional measures, distributions, etc, some of which c.f. has used in his software, to look into that “how.â€

Post Reply