Wednesday, 6 June 2007

Update: Formulas, Walk-Forward Testing

As I mentioned in a few posts, I've been planning to put some formulas up here so readers can create their own COTs database. I had some trouble doing so using Google Docs & Spreadsheets (I guess it's still a little buggie), so until I figure out a better way I'm happy to email a sample spreadsheet to anyone who requests one. Just mail me here:

Also, I've been asked by some readers - including one rather irritable quant (cheer up, "Anonymous"!) - whether I've done any walk-forward testing to eliminate the problem of curve fitting. The answer is yes, and happily, my best backtested setups seem to get the best forward-tested results. In those markets where the forward-tested results are weaker, there tends to be a larger maximum drawdown for the setup. And that means my maximum portfolio allocation is automatically smaller to account for the heightened risk. (Click the "Profit/Loss Results" and "How It Works" links to the right for details.)

I'm constantly re-examining and re-thinking my setups as I go because I'm trading them with my own money. One promising approach I've had is to combine signals from various groups of traders. I'll keep you posted on my progress.


steve said...

Thanks again for this data. I have a few villiage idiot question for you. So, WRT the S&P COT, did you just graph the data against the S&P index, to find your buy and sell (bullish/bearish)points?
Now that you have this data, and your data points, what are you trading? ETF? Options?


Alex Roslin said...

Hi Steve,

Thanks for your message. See the explanatory pages (click the links to the right) for answers to your questions. The short version is I chart the data when I started looking at it - also drawing trendlines - but eventually turned to Excel to backtest combinations of moving averages and standard deviations to find entry and exit points. I'm trading it mostly with ETFs.

Take care,