CL always-in
- Thread starter dom993
- Start date
The key datapoint in the stat analysis is what I call the trend UsefulSize, which is the number of points from a trend-change to the next trend-change, always measured so that a positive value indicates a winning trade if taken "with-trend", and a negative value if that same trade would have been a loss (but, of course, the idea is to fade those trend-changes).
This table sums-up the UsefulSize of all trends, and clearly shows that 1) CL has most often a bias towards fading those trends 2) 2013 was pretty good in that area, much better than the prior 3 years.
This table sums-up the UsefulSize of all trends, and clearly shows that 1) CL has most often a bias towards fading those trends 2) 2013 was pretty good in that area, much better than the prior 3 years.
I am now getting to the two most important statistics generated by this market model: the % of trends that need to be traded "with-trend" (TR), and the % of trends that need to be traded "counter-trend" (CT).
These 2 figures don't add-up to 100%, because any trend which UsefulSize is between -3-tick & + 3-ticks (included) is counted as "Ignore" (not enough room to pay for comms, slippage, and make at least 1-tick profit).
What you will see from the 2 tables attached, is that those % are pretty much identical year over year, and 2013 was no exception.
These 2 figures don't add-up to 100%, because any trend which UsefulSize is between -3-tick & + 3-ticks (included) is counted as "Ignore" (not enough room to pay for comms, slippage, and make at least 1-tick profit).
What you will see from the 2 tables attached, is that those % are pretty much identical year over year, and 2013 was no exception.
This is where those stats from the underlying market model become really useful in assessing the validity of the patterns found through statistical analysis.
A basic MC simulation gives the standard deviation for a set of N coin-flips, with the coined biased 33% then 62%. This is used to measure the separation (in number of standard-deviations) of a pattern win% vs. taking random (direction) entries at each trend-change.
For example, using the 62% likelihood of having a CT-trend, on a set of 3000 samples - corresponding to the sample size of pattern 151, the standard-deviation comes in at 0.885 percent. On the 2007..2012 statistical analysis, this pattern has a win rate of 66.1%, corresponding to a separation of 4.6 standard deviation from random entries ... this is close to 99.9999% confidence that the underlying sample-set is indeed not from a random process.
(which, combined with the sample-size of 3000, is a BIG - no, HUGE - element in favor of CL151, which focuses solely on that pattern).
In the existing patterns, I found 3 with about no separation from random entries on that criteria. And it turns out that 2 of them have absolutely ZERO impact on the system overall performance (well, not zero, but less than +/- 1%).
Another 3 patterns have a separation between 1 & 2 stdev.
Another 6 patterns have a separation between 2 & 3 stdev.
7 patterns are between 3 & 4 stdev.
3 patterns are beyond 4 stdev (700, 800 and 3000 sample-sets).
A basic MC simulation gives the standard deviation for a set of N coin-flips, with the coined biased 33% then 62%. This is used to measure the separation (in number of standard-deviations) of a pattern win% vs. taking random (direction) entries at each trend-change.
For example, using the 62% likelihood of having a CT-trend, on a set of 3000 samples - corresponding to the sample size of pattern 151, the standard-deviation comes in at 0.885 percent. On the 2007..2012 statistical analysis, this pattern has a win rate of 66.1%, corresponding to a separation of 4.6 standard deviation from random entries ... this is close to 99.9999% confidence that the underlying sample-set is indeed not from a random process.
(which, combined with the sample-size of 3000, is a BIG - no, HUGE - element in favor of CL151, which focuses solely on that pattern).
In the existing patterns, I found 3 with about no separation from random entries on that criteria. And it turns out that 2 of them have absolutely ZERO impact on the system overall performance (well, not zero, but less than +/- 1%).
Another 3 patterns have a separation between 1 & 2 stdev.
Another 6 patterns have a separation between 2 & 3 stdev.
7 patterns are between 3 & 4 stdev.
3 patterns are beyond 4 stdev (700, 800 and 3000 sample-sets).
Results for the week ending March, 28th, 2014:
- 3 wins ; 0 losses ; net +2405
The end of last week (Thursday 20th) & the beginning of this week (Monday/Tuesday had some unusual spiky / yo-yo price-action in Crude-Oil. To the contrary, Wednesday was unusually lame, with a tight 65-ticks range for most of the RTH session, only broken-out by 25-ticks from 2:25pm EST. The system did fare pretty well in those circumstances, minimizing the number of trades, and staying on the right side of the market throughout the week.
I am an official NinjaTrader partner as of today, I am sure it will be easy enough to find me in today's New NinjaTrader Partners email
- 3 wins ; 0 losses ; net +2405
The end of last week (Thursday 20th) & the beginning of this week (Monday/Tuesday had some unusual spiky / yo-yo price-action in Crude-Oil. To the contrary, Wednesday was unusually lame, with a tight 65-ticks range for most of the RTH session, only broken-out by 25-ticks from 2:25pm EST. The system did fare pretty well in those circumstances, minimizing the number of trades, and staying on the right side of the market throughout the week.
I am an official NinjaTrader partner as of today, I am sure it will be easy enough to find me in today's New NinjaTrader Partners email
Results for the week ending April 4th, 2014:
- 1 wins ; 5 losses ; -3350
Got stopped-out twice this week, and I finally resolved myself to take a break from trading this system, and instead add a 2nd contract to my trading of CL151 (I am using the alternate trade management for that 2nd contract, which aims are larger wins & total P&L, with a smaller win%).
According to the tracking on RAPACapIntro, the ending P&L is +6938, which includes my 1st CL151 2nd contract win (+195), taking this out it leaves +6743, and removing Jan-Feb 2013 which trades in this account were not CLAlwaysIn, should leave about +6385 for 613 trades.
As a side-note, this system is nowhere close to having failed - as of April-4th, the current version (v51) is down -4085 from its P&L peak, which is "nothing" for this system (well, it is about 1/3 of that version historical max drawdown). But at the same time, the performance of v46 on the last 6 months (since I started trading it on October 15) has been positive but somewhat under my expectations (its current DD is -7620, and my account DD is -6240 vs the January peak), and I want to reflect on that.
In the next couple of months, I intend to get back to the drawing board with the always-in concept, and see if I can improve the overall performance, while keeping only large footprint patterns in the system. I have a number of ideas to explore.
Meanwhile, my trading will be focused on CL151 & CLSelective. CL151 is a single-pattern system, which uses the most stable & 2nd largest footprint pattern from CLAlwaysIn, with some additional price-action filters, and 2 trade-management options. CLSelective uses a totally different market-model, naturally leading to smaller footprint patterns (the largest one has about 1000 samples), but has for it that its statistical analysis covers over 10 years of CL price-action (July 2003 to December 2013).
I will keep this journal open, and report both on the ongoing (simulated from this point) results of v51, as well as on the next iteration of this system.
- 1 wins ; 5 losses ; -3350
Got stopped-out twice this week, and I finally resolved myself to take a break from trading this system, and instead add a 2nd contract to my trading of CL151 (I am using the alternate trade management for that 2nd contract, which aims are larger wins & total P&L, with a smaller win%).
According to the tracking on RAPACapIntro, the ending P&L is +6938, which includes my 1st CL151 2nd contract win (+195), taking this out it leaves +6743, and removing Jan-Feb 2013 which trades in this account were not CLAlwaysIn, should leave about +6385 for 613 trades.
As a side-note, this system is nowhere close to having failed - as of April-4th, the current version (v51) is down -4085 from its P&L peak, which is "nothing" for this system (well, it is about 1/3 of that version historical max drawdown). But at the same time, the performance of v46 on the last 6 months (since I started trading it on October 15) has been positive but somewhat under my expectations (its current DD is -7620, and my account DD is -6240 vs the January peak), and I want to reflect on that.
In the next couple of months, I intend to get back to the drawing board with the always-in concept, and see if I can improve the overall performance, while keeping only large footprint patterns in the system. I have a number of ideas to explore.
Meanwhile, my trading will be focused on CL151 & CLSelective. CL151 is a single-pattern system, which uses the most stable & 2nd largest footprint pattern from CLAlwaysIn, with some additional price-action filters, and 2 trade-management options. CLSelective uses a totally different market-model, naturally leading to smaller footprint patterns (the largest one has about 1000 samples), but has for it that its statistical analysis covers over 10 years of CL price-action (July 2003 to December 2013).
I will keep this journal open, and report both on the ongoing (simulated from this point) results of v51, as well as on the next iteration of this system.
Sounds a bit risky and complicated. This contract can (and does) move .50 in minutes. I personally like the flat position, then jump on an opportunity. I pay close attention to volume at the bid ask and watch for attempts to manipulate market in the resting orders, that's when the action starts. Then you have to get direction right and then you get paid. I trade at around 65-70% accurate and take small profits. This contract can print money for you if you work hard enough. No one on this forum will create an automated money machine.
Interesting comments, will be helpful if you can answer below
1) how long have you successfully traded CL?
2) how many cents are your profit targets like? What do you mean by small profits?
3) how many trades do you do a day?
4) Do you trade only CL?
Dom: hope you are alright me asking this question on your thread.