[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Trading Criteria - actual results



PureBytes Links

Trading Reference Links

Hello List,

I have attached a .csv file that will show the progression of trading logic 
(Signal's) that I have been working with. Record # 1 is the first system I 
created and has about 500 lines of actual code (I place comments on the 
same line as the code). Record # 96 has a hair over 2000 lines of code and 
represents the continued refinement of the same logic . Out of these 96 
different signals I selected 18 Signal's that I liked, and then selected 2 
for further development in C++.

As you can see, after selecting the best 18 of 96 Signals I then applied 
different ratios and drawdown analysis that Trade Station doesn't provide. 
The formula's used were posted in a prior post.

FYI:

1. All trades have more or less the same performance with out of sample 
data as the testing data.
2. Data used was: Data: SP A0 11/29/2000 to 11/20/2001 - Tick Data 
primarily using a 3 minute time chart
3. Out of Sample Data came from data accumulated over the years - primarily 
S&P Comstock, and eSignal.
4. There is no hard coded money lost stop, in other words I did not write a 
exit to get me out of the trade if the position lost 'n' amount of money.
5. The maximum Intra-Day drawdown (as calculated by TS) is not necessarily 
equal to the maximum drawdown used in the analysis techniques for the best 
18 Signals.
6. Multiple positions are not allowed - trades were one lots.
7. In the more detailed analysis of the best 18 the Sharpe Ratio is 
calculated using the .els developed by Bob Fulks
8. Almost all of the indicators used in the logic were developed by myself 
over the years.
9. The logic only uses these indicators as confirmation of the actual price 
action, and as a flag to enter or exit a trade.
10. I use a lot of 2-D array's and loops - as I am looking for certain 
price behavior criteria to be met.

Observation:

The two I have (46, 71) selected for more refinement actually make very few 
trades per day. Both of them typically enter into a trade between 9:50 and 
10:30 and are flat by noon. What I found interesting was that the only time 
constraints I placed on the system was that there could be no trades before 
9:40, and all trades exited by 3:58, yet on a lot of the Signals I coded 
there would only be a few trades a day most of which were in the morning.

I did not place a worst case monetary loss maximum in the system - in other 
words get me out if I lose $2,000 (or whatever amount). What I did code was 
price behavior that conflicted with the type of entry. Using this type of 
"panic" exit is much better to my thinking than a panic exit that is 
defined in some monetary amount. The reason was that in every trade you 
must be willing to allow the trade to "breathe" - or gyrate around. This is 
a totally subjective decision as it relates totally to the amount of heat 
you can take. This reflects the way I actually trade - once in the trade I 
assume that I am wrong and I begin looking for certain price action to 
allow me to exit, and I stay in the trade until I see this price behavior 
(I am wrong) OR I will exit when the market shows me that I was right but 
the momentum has now changed somehow. The price action is different. I 
wanted the logic to trade the same way (or as near as possible) as I do in 
real life.

Objective:

The reason I am posting these results is that I would like to receive any 
comments (good, bad, and ugly) about the performance of these Signals, and 
ask the more experienced System developers which Signals they like and/or 
hate and most importantly why? I am trying to develop a ratio/formula that 
will indicate the probable effectiveness of the logic working with new 
"live" data. In other words it is my thought that it should be possible by 
using the performance results created by"in sample" testing, and the 
performance results created by "out of sample" testing to determine what 
the probability is that the logic will continue to work with new live data. 
These comments might trigger something in my mind that will help me to 
identify what ratio's and relationships to use in creating this new ratio.

In looking at the spreadsheet do you see a ratio that I am using that in 
you opinion is worthless? Or is there one I should include?

Let the comments fly,

John

   

Attachment: Description: "Ol.zip"