Tuesday, July 12, 2005

Et tu, Time Stops?

I'm still here and busy as a bee. I'm currently knee-deep in research for one of my new trading systems. About all I can disclose is that it requires intraday real-time monitoring of symbols. Not a few symbols but a multitude. And that is where I have spent all of my time. Finding and evaluating the products that offer this type of service. Let me tell you...there aren't many.

Hopefully, in the next few weeks I can cover more on this topic.

One last thing I wanted to bring to everyone's attention. Do you use time stops in your systems? I do and if you don't, you should. Time stops get you out of a losing trade or more importantly get you out of a wash trade. You know, the type of trade that just sits there and wastes your capital? Anyway, how do you set your time stops? Do you use trading days or calendar days?

Recently, I had one of my preconceived notions challenged pretty good by one of the great resources over at the Wealth-Lab forums. My opinion has always been that I can control trading days not calendar days. In essence, I can sell the position on a trading day but cannot sell it on all calendar days such as holidays or weekends. So, my stance has always been that in setting up time stops one should count the number of trading days in the trade...not calendar days. Wrong!

The poster let it be known that holding over the weekend is putting my position at risk and thus those 2 days should be counted along with the trading days of the week. And he was so right. And it was one of those Aha! moments when you finally realize something that should have been so simple all along. Anyway, I went back to one of my short-term daily systems and readjusted the time stops to take into effect the weekend. Here's an example...

If my time stop was 6 days and I initiated the position on Monday. Then I should exit at the close of the next Monday if I was using trading days. But, if using calendar days, when should I exit the position? Next Monday? This coming Friday? Good question. And that's what I programmed and tested for. And it turned out that if the day was Friday and my time stop was going to close on Monday then I should close the position on Friday and avoid extending the risk on the position by 2 days.

Do you understand? This probably makes total sense to traders that do not system test their systems. But, for system traders, most system simulations use trading days in determining the number of days a position is held. So, it is natural, to follow that logic in testing your time stops. But, if it wasn't for the great AutoRun product that works with Wealth-Lab that uses calendar days to determine days held for a position...I never would have questioned this very simple piece of logic.

Some may say what difference does it make...just a few days shouldn't make that much difference to a system's performance. And you're right. There's not that much difference. The system I tested with only reduced the system's max drawdown by a few percentage points while increasing the average profit a few percentage points. Also the average days held in a position was reduced by a few days. A few percentage points here and there and it adds up. Kick in a reduced number of days a position is held and more time is given to trade the edge the system was created for in the first place. Trade the system for several years and those little differences begin to grow.

The devil is in the details...and if you're a system developer like me...you spend hours upon hours digging into those details. And wouldn't have it any other way.

Later Trades,



Dan McCarthy said...

Interesting post- as I'm sure you know, all trading systems do their best to estimate some element of reality with simplifying assumptions. In this case by treating all days equally, you make the implicit assumption that the risk/return profile of all days are homogeneous. If you haven't run this test already, it's interesting to compare price volatility on Mondays with price volatility on non-Mondays... or even a day-by-day breakout of price volatility. Banker's trust did this a long time ago and found Monday's to be far more volatile, all else equal... of course that was back when the S&P had a positive serial autocorrelation of .3. Haha.

Keeping track of, managing, knowing the impact of and monitoring all the assumptions we make can make or break any investor- quant or otherwise.

Mike Taylor said...

Dan, excellent feedback. Yes, you are correct, I'm assuming all days are homogeneous. Entirely due to testing of days of week and how they impact my entries in my current system. I even tested other seasonality factors in regard to entry. But, you caught me much like the other fellow on the WL Forum. I failed to test days of week and their impact on my exits and more importantly the impact they play on the variability of my system returns.

I plan to make this my next series of tests. I'll post my results if they uncover anything interesting.

Thanks again for your comments.