Wednesday 11 July 2007

COTs Beat Gold by 99.5%: Setup Details Revealed

I'm going to start posting tables from time to time with a breakdown of signals and results for my setups based on the Commitments of Traders reports. Here are the latest results for gold:

http://spreadsheets.google.com/pub?key=pfhv09UXvNeuZdcDtMtrTKA

This setup works by trading on the same side as the commercial traders when their net percentage-of-open-interest position is two or more standard deviations from its 18-week moving average (using the combined futures and options data).

I use the same parameters in my setup for the HUI Gold Bugs Index, except that setup has much more profitable results. Click "The Trading Setups" link above for more details on this and my gold setup.

Note that the HUI setup scores at the 94-percent confidence level for profitability - less than the 95-percent level I want to see in my setups. This is because we've got less data for HUI than for gold. The HUI data goes back only to 1996, with the first trade signaled in Feb. 1997. With more time and data, I expect this setup will breach the 95-percent confidence level soon.

By pure dumb luck, I think I stumbled across my COTs-based trading system at a good time. It's only recently that there's enough historic data to find robust trading systems that are both highly profitable and can stand up in validation testing—an important process for avoiding the dangers of curve-fitting.

In backtesting, I found plenty of incredibly profitable setups that had a small number of trades. The problem with relying on such setups is it’s impossible to tell if the gains are just coincidental and say anything about future performance. Validation testing can help us overcome this pitfall of system development. Good luck in your trading/investing!

9 comments:

Anonymous said...

Hi Alex:

Regarding the COTs Gold Setup, have you looked at performance by possibly not taking a short position in a "bull market" (defined for example as an upsloping 18 or 40 week moving avergage) but just closing the long position, and waiting for the next buy in a "bull market" and vice-versa for a "bear market"?

In the COTs Gold Setup you could have avoided a couple drawdowns by adding this parameter.

Not sure how it would effect the other markets but might be an interesting exercise how it effects overall return.

Thank you for your blog!

Brian

Alex Roslin said...

Hi Brian,

Thanks for sharing your idea. This is an interesting idea that no doubt could be useful here and in other markets. The only drawback is it adds constraints (trading rules) to the system that would tend to reduce our confidence level in future performance in validation testing. However, in some markets with a great number of trades, that wouldn't have a significant impact.

Another option would be to use the trading rule you've suggested to influence position size.

Regards,
Alex

Anonymous said...

Puzzled why there were no buy or
sell signals in 2006 especially
May and September for sell.

Alex Roslin said...

Hi,

Thanks for your question. The commercial traders were quite bullish in May and Sept. 2006. No setup can be right every single trade or catch every fluctuation. However, that buy signal still ended up being highly profitable.

I'd also refer you to a couple of earlier posts that addressed the issue of drawdowns and catching longer trends:

http://cotstimer.blogspot.com/2007/06/big-money-in-main-movements.html

http://cotstimer.blogspot.com/2007/06/like-casinos-do-it.html

With the setup details I've provided for gold and other markets, I encourage readers to create their own spreadsheets to see what the COTs signaled and why.

Regards,
Alex

Unknown said...

Hi Alex,
I've been working with COT data for many years so your articles on Kitco piqued my interest. I'm impressed by the effectiveness of your approach and I appreciate being able to study your results and methods. I've tried to duplicate your signals but my numbers don't quite match yours and I'm wondering if you can set me straight. In gold for example on 29/5/2007 my net commercials/OI reading was -.2788 (125989-244684)/425688, the 18 week simple moving average of this ratio was -.3877 and the std dev was .0493 (using Excel's StdDev function vs. StdDevP) which means the Ratio was 2.21 sd from the moving average (-.2788 - -.3877)/.0493). My readings for other signal dates you've published show similar differences. I hope I'm not being dense and I appreciate your patience with this long question. However I'm hoping you can offer some suggestions about where my calculations differ from yours. Thank you. Mike

AndrewL. said...

Alex, tried to download cftc to google docs and spreadsheets but to no avail. Also tried to cut and paste into your spread sheet but not much progress even in google groups.Can you point me in the right direction ?

AndrewL. said...

Alex,
Ok, i've got the file to google docs and spreadsheets but it is in csv format. What next ? Would /could you tell me what to do next? I would like to have this figured out before 3:30 Today > Thank you .

Alex Roslin said...

Hi Andrew,

I wouldn't advise doing this through Google docs and spreadsheets. I use that only to publish some documents. Try it in Excel and follow the CFTC website instructions. They're not like Ikea instructions! They actually make sense and are fairly straightforward. Let me know if you still have problems.

Regards,
Alex

Alex Roslin said...

Hi Mike,

You appear to be using different data, probably the futures-only data. This is a common mistake that many readers write in about. I use the combined futures-and-options data. For simplicity's sake, you may also subtract column BA from column AZ.

Best of luck,
Alex