Rich, I understand you are just a messenger... and I was not trying to argue with you either. I appreciate your contribution.
thanks for sharing your system idea rcreal someone wrote "Are you positive the 2pt stop is really working? Sometimes TS has problems with trades that are entered and then exited or stopped on the same bar. " My sentiments as well see below TradeStation makes several "assumptions" as to the way a bar unfolds. In my experience trades with entry and exit within the same bar are unreliable ( ie test out optomistically and unfold very differently when followed realtime) There is a "bouncing ticks" testing parameter that controls some of the software assumptions as to how the bar unfolded If it appears to be working realtime make sure an statistically significant sample is observed. IHO Net
just to test if rcreal's model has any statistical significance, don't use any stop at all and always exit on close. my experience is that the level of stops really doesnt matter for the overall bottom line, therefore, if the model is valid, then statistically the S&P should have a tendency to close towards the direction of the first 2/3 of the day. rcreal, could you remove the stops and test it? Thanks.
Thanks for all the feedback. I've done some light backtesting to discuss the two issues: #1 - The potential problem w/TS on stops. I mainly use TS2000i for testing, however I used a friend's pc to test the same system on TS4 and TS6. Approx. the same trades were taken. For further testing, I manually walked through all tests since 1/22 using Qcharts. From what I can tell, TS seems to handle the 2 point stops fine. The function used is originally from TS4. I've had trouble manually coding stops for the same bar exit. Hence the use of this function. #2 - Stop placement - I went back over the 12/4/2000 to 12/14/2001 sample period trying a 5 point stop and no stop at all. The results were interesting. The 5 point stop caused further losses. No stop was statistically the same as a 2 point stop, however there was one day (9/19/2001) where there was a $7800 losing trade (ouch). The idea of using a daytrading system seems valid. Any system that has a profit factor of 1.5 or greater over a large enough sample period and a sufficient comm/fee $ amount used should be tradeable. The idea of using a trading system which trades intraday only allows the use of daytrading margin further enhancing the leverage futures offer. If anybody so inclines to improve or modify this *possible* system, please let the ET community know. Thanks, Rich
dozu, I concur with your comments. A system that is highly dependent on stop placement is most likely an over-optimized system. My experience has been that a robust system will test out well with no stops, although there may be the occasional wild loss. I like the concept of this system, but I do have some doubts about how a 2 point stop can possibly work on 135 minute bars.
Thanks rcreal for your further testing, since no stop has statistically the same result as a 2-point stop, then I think the system is definitely a good one. I think where to put the stop is then just a personal preference, a tight stop will shake you out more often, but will prevent disasters.... the aggressive type can use looser stops. IF the edge is there, the edge is there no matter where you put the stop. I think the difference in results using a 5-point stop is probably insignificant, as there were only 100ish trades in rcreal's orignal test.. the sample wasn't terribly large.
This is a little off topic, however since another member brought up oddball it seemed like a good place to put it. Mark Brown posts the results of Oddball here: http://www.oddballsystems.com/stats/index.html In speaking w/MB a few months ago, he mentioned that he had an "improved" version of Oddball called Fastball, which took fewer trades but performed better $ wise. MB posts the results of Fastball here: http://www.oddballsystems.com/stats/fastball.html When I spoke w/the commodity shop, the brokerage licensed to trade this system for their clients, a broker mentioned that they are standing around the pc waiting for a signal to trade. The key piece of information he mentioned was 1 o'clock. Something about hte system only trades at 1 o'clock. This got me to thinking ... could modifying the oddball code to only let it trade at 1 pm make a difference? Well, the results were interesting. While my stats differ from MB's, the filter of a 1300 hr trade only trade did reduce the number of trades significantly. Resetting the Buy zone and Sell zone to 0 improved the performace about 10%. Running an optimization from -10 to 10 for each variable turned out matching $ amounts to MB's stats, but the number of trades was off about 10%. When I have some further time, I'll play with this and post my results. I'd be interested in hearing what others might have to say. Rich
using 0 as buy/sell line will probably give us a neutral bias, which is good.... now instead of "always in", lets pick "better" conditions to get into the market, so could you try this out: if position == flat && ROC>2, go long, if position == flat && ROC<-2, go short, if position == long && ROC <0, exit long position if position == short && ROC > 0, exit short position I believe this will reduce the number of trades, and also stay out of the chops when market has no direction and ROC hangs around the 0 line... in other words, when we are flat and ROC does not break either the 2 or -2 threshold, we would stay flat. of course you can run a little optimization on the numbers to use for entry and exit... e.g., instead of 2/-2, use 1.5/-1.5 etc. if rcreal or anybody could test this out and post results here, would be much appreciated. I think this is gonna be interesting. by the way, I suppose MB didn't post the logic of the fast ball, but if it also uses a bias setting like 3/1, then I doubt it's robustness in real trading. Also, the 1pm stuff sounds like optimization to me, there isn't really anything special about 1pm such as bond market open/close, economic number announcement etc....
I'm at home and not at the office, so I've had to use Quote.com data. I've used the SP index vs. futures, also. However, this should give us a close enough comparison. I don't see any significant relation to 1 pm either, however this doesn't allow us to draw a conclusion one way or the other. Deal only in facts ... analyze the data and see what results. The rough tests follow: I'm including the exact code and data used so others can try things out for themselves. One change I made is I used 30 min bars vs. 60 as Oddball trades on natural hours and for some reason, TS2000i would allow plotting of natural hour bars with third party data??? The length of the rate of change was adjusted to 14 to compensate for this. Code for original oddball: Inputs: RL(14), BZ(3), SZ(1); if time > 0930 and time <= 1600 then begin If ROC(Close Data2, RL)>BZ Then Buy; If ROC(Close Data2, RL)<SZ Then Sell; end; Code for oddball-dozu888: Inputs: RL(14), BZ(2), SZ(-2); if time > 0930 and time <= 1600 then begin If marketposition=0 and ROC(Close Data2, RL)>BZ Then Buy; If marketposition=0 and ROC(Close Data2, RL)<SZ Then Sell; if marketposition>0 and ROC(Close Data2, RL)<0 Then exitlong; if marketposition<0 and ROC(Close Data2, RL)>0 Then exitshort; end; Code for oddball-rcreal (1300 hrs filter): Inputs: RL(14), BZ(0), SZ(0); if time = 1300 then begin If ROC(Close Data2, RL)>BZ Then Buy; If ROC(Close Data2, RL)<SZ Then Sell; end; The SP data can be downloaded from: http://www.geocities.com/cgr8deals2000/SP5.zip The Advn issues data can be downloaded from: http://www.geocities.com/cgr8deals2000/adv5.zip Time period tested was 12/4/2000 to 12/14/2001. Results of this (crude) test follow: