I identified about a year ago a reversal pattern on CL, which seemed interesting enough for me to start developing an automated system for it. I won't disclose the specifics of the strategy here, but the idea is to look for for a trend reversal manifesting itself by price making a HH directly off a LL (reversal off a down-trend) - the entry long is in the 1st pullback following that HH (reverse everything for a short after a reversal off an up-trend). I use 2 targets, target-1 is "scalping" and has an average reward of about 85% the average risk, target-2 is "runner" and has an average reward of ~250% the average risk. This reversal pattern does show-up in all timeframes, and I chose to go as "low" as possible in order to maximize the number of opportunities. I decided to work off a 200-volume chart (CL), which initially seemed a good comprise (I would have used a 100-volume chart, if it wasn't for a limitation of my trading platform which limits the number of bars in a chart to 65,000 - CL can actually go higher than that on a single month on a 100-vol chart). The real key to this system is pivots identification. I developed a 1st version that I took live mid-September 2011, only to stop trading it early November as the system experienced a severe loss of performance (which really had started early August). That loss of performance did continue into the new year, until around the start of active trading of CLH12 (late January). Meanwhile, I was frantically scrutinizing the charts, looking for clues. But really, I couldn't find any "reason" for that loss of performance, aside from using "bad" pivots. I then reworked the pivots identification mechanism ... that yielded a significant improvement, manifesting itself not only in the last few months. But it remains that the last quarter of 2011 had the most severe drawdown period of the backtesting history (I currently use tick data since October 2009 for this backtesting, this is 28 months, total number of trades in backtesting is 554). Attached is the P&L curve out of the latest backtesting. Interestingly, the system recovery -since the beginning of CLH12- is only forward testing. I am quite concerned by the amount & the duration of the recent drawdown period. I did MC sim using the trade distribution up to the start of the drawdown, for a number of trades corresponding to what happened since (108 trades) ... the 7,000 drawdown is 4 std-dev away from the mean, and the drawdown duration itself (96 trades) is 8 std-dev away from the mean. But I really lack experience to interpret these figures, and I would appreciate your comments/feedback. To answer some questions I know will be asked: - the system is all about price action ... it uses no indicator whatsoever, only H/L pivots & projection techniques from these - system accounts for $5 comms per contract / trade - system assumes 1-tick slippage on all stop exits - system entries & targets are LMT, assumed filled only when price trades through the limit - system P/F = 1.95 on entire 28 months ; it is down from 2.25 on the 1st 24 months of the backtesting period - system P/F for target-1 is 1.65 on entire 28 months - system P/F for target-2 is 2.00 on entire 28 months
edges come and edges go..in fact, you most likely didn't have an edge to begin with...just a selection bias or curve-fit go back to the drawing board and develop something else.... move along..move along people...nothing else to see here..
Hi, Here you have some more questions Did you run it trough an optimizer? Did you test it out of sample before forward testing?
No optimizer, but a lot of manual testing to find out the best settings. I reworked the pivots based on observations made in the drawdown months ... so in a sense, the prior 23 months could be considered as out-of-sample from that point of view, but certainly not from a basic system point of view. The following figures are month by month difference in P&L between the new version & the old one: -970 4150 -1320 -1090 -210 510 410 820 -820 -570 590 -210 380 40 1320 -920 6240 -2080 -2930 860 -40 -110 1630 3860 820 2360 3210 1770
Ummm, if you are using the traditional pivot point calculations for S1, S2, R1, and R2 for Weekly and Monthly levels my advice would be for you to check your work again - especially if you are fading levels based on the trading day's opening mark and getting flat by that day's close. You also need to consider your max daily drawdowns, and your largest winning streak versus losing streak. Also, look at how fast the 20 day historical vol and the average trading range expanded during your sampling period - it is difficult for an on-the-run range study like Pivots to cope so quickly, and your drawdowns will reflect that ( bigtime ). My 2 cents, thanks for sharing and good luck with it. All technical studies have real limitations, so don't be terribly discouraged - and there are ways to cope.