acrary, thank you for your work on the filters. the only concern i got is the reduction in the number of trades. I had 1422 trades; you reduced it to 202. This is a big reduction in the 'trades sample size'. When I got some results over 1422 trades, I feel more confidence in them than if they were over 202 trades. Please correct me if I'm wrong.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
working system, needs improvement
- Thread starter traderkay
- Start date
Quote from Walther:
To acrary : what if you test it with 20 or 15 points stop? Average loss of 1,700.00 seems to be too large.
Walter
I should have noted the testing was done on the SP market and not ES. The $1700 ave. loss on the short trades was about 7 points or about $340 on the ES. I used 20 points because it was near the average daily trading range and all I was doing was trying to avoid the outlier losers.
With the max. loss reduced to 15 points, the average loss goes up above $1,800 on the SP or about $360 on the ES.
Here's the summary using 15 points for max. risk per-trade.
Attachments
Quote from traderkay:
acrary, thank you for your work on the filters. the only concern i got is the reduction in the number of trades. I had 1422 trades; you reduced it to 202. This is a big reduction in the 'trades sample size'. When I got some results over 1422 trades, I feel more confidence in them than if they were over 202 trades. Please correct me if I'm wrong.
Yes, I'd love to have more trades. However, the difference between 200 and 1400 is statistically insignificant. As is, the sample size of 202 with a mean of +404 and a std. deviation of 1313 has a confidence level of 95% that the average trade is 404 +/- 183.
To prove 404 is the expected average in the future at a 99% confidence level would require over a million trades.
The number of trades can be a illusion. Afterall, it takes over 15,000 flips of a coin to have a 99% confidence level that the next flip will have a 50% chance of being heads.
I don't know if this has any edge as I haven't converted my daily stuff to intraday yet. It'll be interesting to see how this compares to random entires to see if I'm curve fitting or finding a potential edge. If it is an edge, the sample size is fine. I've implemented systems with a sample size of less than 100 that were so far outside of random that I couldn't ignore them.
The other thing that gives me some confidence is the sharpe ratio is 2.19 in the test. My experience has been that tests with a sharpe ratio above 1.6 are worth pursuing.
At any rate this has helped me figure out what I need to do to build a intraday testing suite. I've learned alot and have a long todo list.
Good luck with whatever you do with this.
T-kay
If you look at to posts you can see this.
Item 5 in your response in the other thread you originated says that you see 3 to 4 trades in the ES a month.
I suggested elsewhere that there are usually 3 to 5 trades a day. In a specific example i gave for last friday (14th), I noted 4 specific trades for the day that looked like a total of 45 ES units to the left of the decimal point. capturing 22 of them seems a reasonable level to be operating at.
This second paragraph happens every day as it turns out.
If there are an average of four per day and your see 3 or 4 a month then you are "bridging" across or skipping over many trades by using what you do.
People commonly look at markets and determine how to make money and then look more closely to see how to optimize their capital appreciation.
The elements of the compound interest formula drive this as you would expect.
I view about seven levels of market duration. Using these seven levels you can determine the average number of trades per year and the average profit per trade. By using the same initial capital as the remaining variable, it quickly becomes clear what the market is telling you to focus upon to make money.
To make money the essential features are utilizing the portion of the market cycle that maximizes a trade and minimizes the duration to get the optimum. Because ES is a singlton market, there is no advantage to shortening the cycle. In trading equities there is an advantage because the market consists of a large universe to be rotated through. There you enter late and exit early because you are seeking money velocity crossovers for switching from equitiy to equity.
You enter late on ES trades because of your restriction and the fractal you have chosen. Both are things the market chooses and you have abrogated that opportunity for the market.
You also need to recognize that the pace of the market changes and you need to change with that to a suitable monitoring fractal.
Your long MA is the principal money making "killer' in your set up. the signal in itself is a "killer" too for any MA.
By linking the protection (which is not for making money at all) to first of all exits and secondly to targets, you again take over the function of the market.
If you look at to posts you can see this.
Item 5 in your response in the other thread you originated says that you see 3 to 4 trades in the ES a month.
I suggested elsewhere that there are usually 3 to 5 trades a day. In a specific example i gave for last friday (14th), I noted 4 specific trades for the day that looked like a total of 45 ES units to the left of the decimal point. capturing 22 of them seems a reasonable level to be operating at.
This second paragraph happens every day as it turns out.
If there are an average of four per day and your see 3 or 4 a month then you are "bridging" across or skipping over many trades by using what you do.
People commonly look at markets and determine how to make money and then look more closely to see how to optimize their capital appreciation.
The elements of the compound interest formula drive this as you would expect.
I view about seven levels of market duration. Using these seven levels you can determine the average number of trades per year and the average profit per trade. By using the same initial capital as the remaining variable, it quickly becomes clear what the market is telling you to focus upon to make money.
To make money the essential features are utilizing the portion of the market cycle that maximizes a trade and minimizes the duration to get the optimum. Because ES is a singlton market, there is no advantage to shortening the cycle. In trading equities there is an advantage because the market consists of a large universe to be rotated through. There you enter late and exit early because you are seeking money velocity crossovers for switching from equitiy to equity.
You enter late on ES trades because of your restriction and the fractal you have chosen. Both are things the market chooses and you have abrogated that opportunity for the market.
You also need to recognize that the pace of the market changes and you need to change with that to a suitable monitoring fractal.
Your long MA is the principal money making "killer' in your set up. the signal in itself is a "killer" too for any MA.
By linking the protection (which is not for making money at all) to first of all exits and secondly to targets, you again take over the function of the market.
It will get fairly clear after a while that primary difficulties have to be remedied before any analysis can commence. most analysis in this vein is not going to make more money simply because it does not deal with the major issues.
If any one has a paradigm yielding 3 to 4 trades a month in an index there is no statistical analysis that is going to make any headway determining either what is right nor what is wrong.
If any one has a paradigm yielding 3 to 4 trades a month in an index there is no statistical analysis that is going to make any headway determining either what is right nor what is wrong.
More or less, I can do my own testing. Although as can prolly be evidenced from my code, I'm far from being a star coder.
Quote from Walther:
Quote from traderkay:
I see, Walther. Any suggestions for filters, that are better than 200MA (except time filters)? Do you think it's possible to achieve 80% without going into rocket science like cycle analysis, DSP, etc.?
Absolutely. Can you do your own testing or someone else is doing it for you ?
Walter
Very good explanation, acrary. 95% confidence level is good enough for me. But it seems that the primary "worthiness" test you would perform is the edge testing which is as I understand, something like comparing the system output with random entries output. I would be very interested in learning how to conduct that. It seems that you were going to describe the process some time ago in another thread but stopped because of noise from other posters. Which is too bad, I hope you changed your mind, and will describe this edge testing methodology.
Quote from traderkay:
I hope you changed your mind, and will describe this edge testing methodology.
I did post the methodology here:
http://www.elitetrader.com/vb/showt...test#post134947