Monday, 12 January 2009

Stop Levels Tweaked

I've made a change in how I calculate my stops for my trading setups based on the Commitments of Traders reports. My existing method of basing the stops on the largest past drawdown in the backtesting seemed a little unsatisfying - besides leading to the potential of huge losses, which never makes me comfortable, even if the position size is commesurately small. Instead, I'll be now using a stop based on the setup's average profit of all the trades minus two standard deviations of the trades results. The stop levels for the two new setups I've developed so far are posted on my latest signals page. (You'll notice all the other setups have no stop levels or portfolio allocations. That's because I'm not trading them right now as I'm retesting all my setups with detrended price data and the Monte Carlo test.) My maximum portfolio allocations will be adjusted to these new stop levels (i.e., never risking a loss of more than two percent of total assets in any single trade). Also, I may sometimes scale into positions by starting off with half of the maximum allocation on the entry date, then purchasing the second half one week later if the trade is profitable. Good luck this week.

2 comments:

Anonymous said...

Alex,

On the new stop loss setup: aren't you biasing the stop limit based upon your own trading history which is a small sample instead of a larger sample such as the individual underlyings trading history? Wouldn't it be more effective to take the last 20 days atr and then look for std dev moves beyond the largest up and down move?

Tim

Alex Roslin said...

Hi Tim,

Thanks for your message. There's always a question about how much data to include. Robert Pardo talks about the importance of having at least one of each of the major conditions: bear, bull and trading range. I think the 1995-today period covers those.

Regarding ATR, I use it in discretionary trading to help figure out stop levels, but I'd prefer a stop based on the volatility of the trading setup's returns, rather than the price. The goal would be to capture as much as possible of the setup's typical action.

My testing so far hasn't produced any decent methods of arriving at stops based on the price action - though I haven't tested the specific idea you suggest.

I'm sure there are many other ways of setting stop levels for this system.

Regards,
Alex