If it keeps the position as is, this seems kind of counterintuitive depending on your exit rules.
CL always-in
- Thread starter dom993
- Start date
Results for the week ending September, 13th:
- 9 wins / 4 losses ; net -385.
At least the win% this week is back above its long-term average. The system was caught long overnight Monday, resulting in a -2500 loss.
CL has had an abundance of abrupt overnight moves lately, where overnight in general has a statistical bias towards chop, and the system is clearly suffering from that current state of affairs.
Using stops would have clearly benefited 2013 so far, despite being consistently inferior in backtest. I am almost at the decision point to turn stops on, at least overnight.
- 9 wins / 4 losses ; net -385.
At least the win% this week is back above its long-term average. The system was caught long overnight Monday, resulting in a -2500 loss.
CL has had an abundance of abrupt overnight moves lately, where overnight in general has a statistical bias towards chop, and the system is clearly suffering from that current state of affairs.
Using stops would have clearly benefited 2013 so far, despite being consistently inferior in backtest. I am almost at the decision point to turn stops on, at least overnight.
Hey dom do your patterns work on other instruments or just CL?
The current version of the system uses a trend-identification mechanism which is fine-tuned to CL. The patterns are also fine-tuned to CL.
I did test on other markets using a more general trend-identification mechanism, in general the patterns are different market to market (even though, I found one working the same in 6E & CL). ES & YM very close results, but NQ / TF require different thresholds.
I didn't find a one-size-fits-all (markets) pattern / system.
I did test on other markets using a more general trend-identification mechanism, in general the patterns are different market to market (even though, I found one working the same in 6E & CL). ES & YM very close results, but NQ / TF require different thresholds.
I didn't find a one-size-fits-all (markets) pattern / system.
Yo, what happen?
I have been very busy for the last couple of months, and only updated the RAPA website with the system's live results.
That said, the system has had mostly losing weeks since the beginning of July, and as of Monday this week my account was back to break-even, with 24 losing trades in the last 31 trades.
Even though my system stop was not reached, I felt I would suffer more of the same if I kept trading it as is, and given the pace of losses (-12k for just the last 4 weeks) that system stop could be hit in a couple of weeks, or less, and there was no point continuing in the same fashion.
Over Monday/Tuesday I took a clean-sheet approach, and did the statistical analysis all over again, paying special attention to not mix variables in the patterns definition, and to not define the patterns narrowly - all in all, trying to *not* over-fit the system.
The other thing that I did, was to decide on using a reasonable size stop at all times, even when the performance in backtest is about 20% lower that way - but I have learned over the past 3 months, that stops are just there to assist when the market goes out of the usual for an extended period of time.
That new version yields on the 6 years backtest only 60% of the P&L of the prior one for about the same number of trades, but it's forward testing performance on 2013 is in-line with the last 3 years (2008 & 2009 are way above the average, and 2007 well under the average), and the current drawdown only started a month ago, and is of lower magnitude than with any prior version (it is currently quite less than the max historical drawdown).
All in all, I think this new version has better chances of long-term success, so I decided to switch to it, which I did at 2:30am on Wednesday. The first couple of days with it are certainly encouraging, with 5 wins on Wednesday, 1 loss + 1 win today, up +3185 for these 2 days, while the prior version had 2 more losing days.
That said, the system has had mostly losing weeks since the beginning of July, and as of Monday this week my account was back to break-even, with 24 losing trades in the last 31 trades.
Even though my system stop was not reached, I felt I would suffer more of the same if I kept trading it as is, and given the pace of losses (-12k for just the last 4 weeks) that system stop could be hit in a couple of weeks, or less, and there was no point continuing in the same fashion.
Over Monday/Tuesday I took a clean-sheet approach, and did the statistical analysis all over again, paying special attention to not mix variables in the patterns definition, and to not define the patterns narrowly - all in all, trying to *not* over-fit the system.
The other thing that I did, was to decide on using a reasonable size stop at all times, even when the performance in backtest is about 20% lower that way - but I have learned over the past 3 months, that stops are just there to assist when the market goes out of the usual for an extended period of time.
That new version yields on the 6 years backtest only 60% of the P&L of the prior one for about the same number of trades, but it's forward testing performance on 2013 is in-line with the last 3 years (2008 & 2009 are way above the average, and 2007 well under the average), and the current drawdown only started a month ago, and is of lower magnitude than with any prior version (it is currently quite less than the max historical drawdown).
All in all, I think this new version has better chances of long-term success, so I decided to switch to it, which I did at 2:30am on Wednesday. The first couple of days with it are certainly encouraging, with 5 wins on Wednesday, 1 loss + 1 win today, up +3185 for these 2 days, while the prior version had 2 more losing days.
Cool DOM, thanks for the update. If you dont mind my asking, what kind of profit factor are you getting in backtest? Also, what kind of profit factor in forward test?