Brian, Thanks for your comprehensive response! I am in the beginning of my system development carrier so I am trying to get a grip of the whole area. I have looked at MCS but not done anything about it yet.
Back to my problem with k-ratio. I just learnt there are now 2 versions of k-ratio proposed by Mr. Kestner. The original one from 1996 and a new and improved one in 2003. AB in its current version implements the second version. In the second version, one of the parameters is the number of bars used in the back testing. So, I assume that one no longer can say that a k-ratio above 1 is "good"? It can only be used to compare different systems tested on the same dataset, or? Also, to get the new k-ratio, you divide the old k-ratio with the square root of the number of bars which means that the new k-ratio will be a smaller number so again, "1" perhaps does not mean so much any longer? Am I correct in my above observations? Kind regards, /Fredrik --- In [email protected], "brpnw1" <[EMAIL PROTECTED]> wrote: > > Fredrik, > > I've seen K Ratios in my own testing, that have approached "1" and > beyond. > > I used to try to use all the different available ratios and look for > a high-point commonality amongst all of them, in the 3D optimization > graphs. Some of these you will have to look at the numbers outside > of AB, since AB likely does not support them. I created my own Excel > spreadsheets for doing this, way back when. > > What's proven far, far more useful for my system testing, though, is > Monte Carlo Simulation. I would focus far more on MCS, than on any > other aspect of system testing. Start there if you haven't already. > If you get decent results with MCS, you really don't need to > optimize your system anymore, IMO. This is the same method the > better, more sophisticated financial advisors tend to use. > > I build systems that work, just by visually looking at them. > Optimizing them would be a waste of time, considering I can't > optimize my own body and mind (one can only compensate for > weaknesses with trading methods such as scaled trading), and I'm the > largest part of the trading equation. > > Now if your goal is to build a "black box" trading system that does > not involve human intervention, you will want to make it neural so > that it intelligently optimizes itself on an ongoing basis. Side > note: if would be wonderful if AB eventually integrated MCS > functionality. > > From my system development notes: > > d. ROI ratio = ( net profit / maximum trade drawdown ) > i. If comparing ROI ratio between systems, must use same > historical data for each system test. Look for highest ROI ratio. > e. Sharpe Ratio = measure of risk adjusted ROI. > i. Above 1.0 is good, above 2.0 is very good, 3.0 is excellent > ii. Decreasing ratio means system losses hurt more than gains > help > f. K-Ratio = Detects inconsistency in returns. Measures both > profitability and consistency of returns and then returns them in > one single number. > i. Should be 1.0 or more. > ii. Ratio dependent on length of historical data used. > g. Pessimistic Return to Risk Ratio (PRRR) = Large positive returns > are not penalized like with the Sharpe Ratio. > i. ADD = avg daily drawdown, AWT = avg winning trade, ALT = avg > losing trade, Wins = number of winning trades, Losers = number of > losing trades > ii. PessimisticNet = AWT*(Wins-sqrt(Wins))-ALT*(Losers+sqrt > (Losers) > iii. PRRR = PessimisticNet/ADD > > B. Optimize each indicator against selected stock historical (EOD > and weekly, separately) data. Do not use all available data since > building a system that works on all data is not practical (too time > consuming to build). Do random sampling to optimize from, and then > also predict other random historical data. Avoid curve fitting!!! > a. COMPLETE SINGLE BULL-BEAR CYCLES > i. In Sample = 4/1/2002 - 2/1/2004 > ii. In Sample = 11/1/1998 - 10/1/2001 > b. ULTRA-BEAR CYCLES > i. In Sample = 10/1/2000 - 10/1/2002 > c. ULTRA-BULL CYCLES > i. In Sample = 1/1/1996 - 4/1/2000 > ii. In Sample = 4/1/2003 - 2/1/2006 > > If you get decent MCS results in the worst market conditions, you > system will probably work in any conditions. One hint: momentum. > > Brian > > > > --- In [email protected], "broman1003" > <fredrik_broman1003@> wrote: > > > > Hi, > > > > I am wondering about the implementation of k-ratio in AB. In other > > software products I have used, for a good system, k-ratio normaly > is a > > number above 1. And according to AB manual, it should be a number > over > > 1. However, when I backtest systems in AB, I have so far never > seen a > > k-ratio over 1. > > > > So, is there a problem with the implementation or is it me who is > at > > fault? > > > > Kind regards, > > /Fredrik > > > Please note that this group is for discussion between users only. To get support from AmiBroker please send an e-mail directly to SUPPORT {at} amibroker.com For other support material please check also: http://www.amibroker.com/support.html Yahoo! Groups Links <*> To visit your group on the web, go to: http://groups.yahoo.com/group/amibroker/ <*> Your email settings: Individual Email | Traditional <*> To change settings online go to: http://groups.yahoo.com/group/amibroker/join (Yahoo! ID required) <*> To change settings via email: mailto:[EMAIL PROTECTED] mailto:[EMAIL PROTECTED] <*> To unsubscribe from this group, send an email to: [EMAIL PROTECTED] <*> Your use of Yahoo! Groups is subject to: http://docs.yahoo.com/info/terms/
