Monday 10 March 2008

Thinking About Robert Pardo's New Book

Not long ago I got a copy of The Evaluation and Optimization of Trading Strategies, Robert Pardo's revised version of his classic 1991 book on trading system development, Design, Testing and Optimization of Trading Systems. It's giving me lots of ideas on how to optimize my trading system based on the Commitments of Traders reports. I've already implemented a couple of basic concepts, like limiting testing to setups with at least 25 trades and excluding setups that use a moving average period of more than 35 to 45 weeks. (The latter is because a setup won't be reliable if it uses more than 10 percent of the available degrees of freedom based on the total weeks in the dataset.) But my main goal is to develop some walk-forward testing for my system based on Pardo's ideas. Stay tuned. Highly recommended book!

11 comments:

Anonymous said...

You should really implement an Amazon link to your book recommendations within the article. It helps me checking for more infos about the book one click away and it helps you making some cash from your site traffic. I would honestly grant you some cash for all your work.

Alex Roslin said...

Good idea - thanks. I'll look into it.

Regards,
Alex

Anonymous said...

Alex - have read some of Pardo & Art Collin's work.

Imagine they would be nervous about the widely different moving average and standard deviations you are using for the different instruments you trade.

Even if you use fixed paramter values for Comercials, another for Small Traders etc that would reduce the chance that you are just [unwittingly] backfitting / data mining.

Paul

Alex Roslin said...

Hi Paul,

I'm not sure. Haven't seen anything in the book so far on that score, but I'm still far from finished. The WFA seems like an excellent method for testing the robustness of the overall strategy, but my guess - based on how well the COTs data gives signals for prices - is that the walk-forward efficiency will vary for various markets. I will have to give it more thought as I go forward.

Regards,
Alex

Anonymous said...

Alex, here is Art Collin's system
http://www.tigersharktrading.com/articles/11539/1/The-Odds-Czar-Index-Futures-Biases-for-March-13/Page1.html
Note he uses the same MA for all indexes and currencies.
[He used to use an optimised one for each and got creamed]

Here is his book [foreword by Pardo]
http://www.amazon.com/gp/product/0470038659?ie=UTF8&tag=tigersharktra-
The whole story of the system developement is in there.

He also welcomes email contact, especially from other system developers

Paul

Alex Roslin said...

Hi Paul,

Thanks. Without being familiar with what he did originally and in the new system, I can't comment on his getting creamed, but I will certainly check it out. However, my understanding of WFA thus far suggests to me it tests the overall robustness of the trading strategy, not any specific parameter values.

Thus, it leaves open the possibility of changing parameter values as the system is reoptimized regularly every so often, but presumably also different parameter values in an entirely different market.

In other words, the question would be, Why couldn't you optimize parameter values for a different market if you can do so for a new timeframe in the same market that reflects new market conditions?

Curtis Faith's excellent book Way of the Turtle cautions against doing so only because optimizing for each market may not leave enough trades to satisfy statistical robustness conditions. But the COTs datasets seem to provide more than enough trades (100+ in most of my recently revised setups).

However, I'm no stats expert, so clarifications are welcome.

Regards,
Alex

Anonymous said...

You know a LOT more about stats than me Alex.

My contribution is soley to point you twoards Art Collins work cause he is a colleague of Pardoe's.

And his story [regarding moving average use] is intersting, especially as that is a big part of your work.

He can barely use Excel, so he is really not into stats. He uses system developer software.

Paulkij

Alex Roslin said...

Hi Paul,

I appreciate your bringing his work to my attention. It does look quite interesting. Hope to study it more closely.

Best regards,
Alex

Anonymous said...

Hi Alex - me again :o)

Trying to reproduce your Gold setup in Excel, without success.

Am using CMX Gold, Non-Comercials [Columns AW, AX] Futures & Options, ma=5, sd = -1.2

My s/s gives a Buy on 22 Jan [data date], and stays there to date [previous signal was a one-week Sell]

Any ideas what I am doing wrong ?

Paul

Anonymous said...

Think I got it Alex

Format I was using was 2 weeks behind.

Will keep an eye on your Gold signals to check

Paul

Alex Roslin said...

Hi again,

In response to Paul's earlier comment (not sure if that's the same Paul as you) up above about using different parameter values for each market, I just found a passage in Robert Pardo's latest book that talks about this very issue. He says he advises doing just this. I'll put up a blog post on this important subject with more details. Thanks for raising it.

Best regards,
Alex