Quote from jo0477:
No one is saying that its not possible to code a rule based system based on "contextual clues", but I believe that it must be extremely difficult. I would imagine the problem would lie in culling the logic down to a level of complexity where it is codable yet keeping the system functional and at a point where you are comfortable letting it run. Congrats Nodoji if you managed to accomplish that because I'm sure it was a tough job to get there!
When I was working the energy desk on the west coast a few years back, I was lucky enough to meet a guy who did a bunch of mean reversion stuff on the side and he let me work with him once in a while... way over my head, but I did pick up enough on the programming end to get a feel for the time and effort it would take to put a quant strategy together. Difference is that type of system is black and white, either it worked or not. For example: if the system took 100 trades and they were all losers - there is no "qualifying" of trades. If the code was functioning properly, then the strategy was flawed - all trades qualified. If the code was broken, all trades were disqualified and you went back to testing.
Now assume you are a manual trader working on a system, trying to determine if its viable. You take 100 trades according to various rules. 50 winners and 50 losers to breakeven. Now the subjective part comes into play and this is where I assume most get into trouble? Going back over your trades, maybe you decide all 50 losses were executed according to your rules and count them towards your statistical analysis of the system - its a scratch. Or maybe 10 of the winners were poor trades so they should be excluded - now its a loser. Lastly, maybe 10 of the losers shouldn't have been taken - now its a winner.
My point is, a statistical analysis is only so scientific if the data being analyzed is being subjectively selected by an individual person. How do you guys manage to avoid sort of curve fitting your results and get your rule base to the point where you are able to automate successfully (since I see several of you have apparently done so). I would really like to have some guidance here if I ever do try and attempt it in the future.
And seriously, get a grip Jack. Annual Hedge fund returns in minutes?? It used to only be 4x the daily range, now you're really getting greedy...
I used market surfers posted list. Most were losers and I recognized that the value could have been a winner for me on my system technique.
I looked at how long my current market context was happening and jotted down the times. Then I posted my post.
Consider the percent of my all my capital I have in the market. The numerator is what is making money; the denominator is my total capital. The percent is 94%.
In ES, I use 1500 dollars per contract. On any information display, this is represented by 30 points since a point is 50 dollars.
Surf posted annual percents; most were single digits. One was 20%. 20% in PVT stock trading is two turns where a turn could be as long as 5 days. In PVT trading three things are used: An ATS Universe, 5 rules for trading using volume as a leading indicator of price; and an account which applied 94% of total capital.
7,000 stocks are filtered to 100 stocks to get a reliable universe. For any day, the available HOT list is ranked by performance as part of the list construction. This is a very precise and diciplined set of hoops to get through.
therefore, over a year 6 doublings of capital occur. The captal sequence is: 1>>2>>>4>>8>>>16>>>32>>>64. As a percentage this is 6400%
20% divided by 6400% is the portion of 240, 6 and 1/2 hour days that, pro rata, would be consumed. I was liberal in my prior post.
AS you say you have watched people work to develop skills and knowledge. Some day, you may meet others like Donna or me.
We do GTM's daily for up to bar 30 approximately; today that would be 6 profit takings on six trends on the 5 minute time frame. The opening range was traversed twice; the range expansion was traversed. then half ot of days range was half traversed three times. An entry and five reversals, all event based. Thus the AM trading was a multiple of the ATR; the PM trading will be the same.
To me, surf's annual retruns of losing HF's he thinks ae going out of business are a joke. People invest in these jokes. Surf introduces people to these HF's
I trade stocks using three MLR's of relative duration. The humorous name for this angular velocity comparison and signal generator is "pinwheel trading". We created it as a joke.
How high velocity trading works is having a system that names every bar, bar-by-bar. The names are precisely and uniquely defined. Further the names fall into three sets of 10, 11 and 35. 10 apply to price, 11 apply to volume for trend analysis.
A two HS in a complete algorithm with a PM in the form of a vector, define the basis for the naming of all 56 elements.
Trends, as you may know have beginnings and ends. The list of names of these is 35 elements long. 10 subsets are named to classify these 35 unique ends.
A complete mathematical system emerges that bar-by-bar fully and wholly descripes the market at any moment of its operation.
There are no noise, no flaws or anomalies.
You , as you say, watch people work, This empowers you, you think, to tell me to get a grip.
As I trade I have feelings of comfort, support and confidence. The reason is, is that I always "know that I know.
I have a log that has columns. there I perform a test procedure. My log has the columns of names in their perfect relationship.
I have 10 tables of the subsets of the End Effects; each is mathematically defined.
I know to do the procedure to get permission to do the test procedure. I know if the test procedure fails to yield a rsult of H1 of the agorithm, that then I do the work to anayze for the H2 End effect. since the 35 End Effects are complete and each uniquely defined, then I know I know to reverse to take profits of a completed trend and then to be positioned to begin to take profits in the ensuing new trend.
I have all of these items on five sheets of paper;three for End Effects and one for a log and one for permission to use the test procedure and the test procedure.
How long does it take to be able to use the sheets flawlessly. About 40 hours of what you say you watch people do.
Do you want me to post the five sheets. No, you do not. You know the reason why; so does surf.
Now you know quant strategies do not work and I have explained why they do not work.
