It has been a while since I posted this, mainly because there haven’t been many trades lately. The other reason is that I’m just tired of writing about it.
Jeremy is putting the finishing touches on his back-end infrastructure so that the Power Dip will run on auto-pilot, tracking the performance on all trades, and updating the subscriber every evening as to new trades and updating profit targets for existing trades.
Yes, the system is in a drawdown, and yes, it is a tad embarrassing to be launching a strategy that is under-performing. However, I believe the strategy is robust, and honestly, the best time to allocate capital, whether it is with a new fund, new manager, new strategy, etc., is during a period of under-performance. Study after study shows how investors typically chase performance, and bail on a manager, fund, or strategy at the exact wrong time. Similar to trading, many investors/traders have the tendency to buy high and sell low when allocating capital.
The system is long Compellent Technologies, Inc. [[CML]] and MasterCard Incorporated [[MA]] .
There are no new buys or sells for Wednesday.
Sounds good, Wood – looking forward to it!
I’m sure you have an answer to this. How do you know when your system is in an expected drawdown, versus when it’s broke?
And at what point are you willing to suspend the trading of a system? What has to happen? I’ve heard figures of a 1.5 standard deviation drawdown from backtesting profiles, but I’m not sure where I come down on the topic.
Have you considered deploying a kernel regression algorithm to determine what sort of trendiness and volatility environment Power Dip performs best in? I’ve got that project on the drawing board. Instead of always trading a particular system in every market environment, the idea is to restrict or expand it based on current market conditions. Somebody used the analogy (Trading Recipes dude) that a super tricked out snowmobile is great in the snow, but doesn’t work so well in the jungle.
Since you trade a large basket of stocks, you can probably find a market (stock) that is particularly suited to Power Dip, which sounds like a countertrend system. It would be interesting to filter out markets that are not trending (range bound) for this sort of system, no?
Milk, take a look at the link below. This is my primary tool to evaluate a system’s health. In order to consider a system broken, I would be looking for months of under-performance that exceeds in both duration and degree the previous drawdowns or phases of under-performance.
In the future, I will be exploring the use of a chi^2 of the win % as an additional tool.
http://ibankcoin.com/woodshedderblog/2009/11/01/is-your-stock-trading-system-sick-take-it-to-the-doctor/
I have read about kernel regression, but do not full understand it. I will look forward to your exploration of that topic.
The system already employs a filter that does pretty much what you describe.
Nice link, as it demonstrates a thoughtful process. Did you come up with the 20Trade average yourself? Also, a good analysis of why you suspect the system is reaching higher than historical drawdown.
It appears Power Dip chooses candidates, which appears to obviate the need for a separate kernel regression filter. I need to give this some thought.
I’m just getting exposed to kernel regression and am reading John Wolberg’s Expert Trading Systems as a reference. Also, Murray has used this book as a reference to create his own kernel regression algo in TradersStudio. The basic idea is to create a neural network that turns a system on or off, or perhaps feeds it or chokes it. If you can feed it trend-relevant data and volatility-relevant data, it may be able to identify when a system performs well, and then apply that output to a trade signal. Instead of taking absolutely every signal, you take every signal that has gotten past the kernel regression sentry. The NN would need to learn under what trend/vol conditions your system works best, which seems fairly straightforward.
My new motto: System trading is measured with a micrometer, marked with a grease pencil and cut with an axe.
CML is killing it today…Hopefully I didn’t just jinx it.