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

[amibroker] Re: Analyaing a system



PureBytes Links

Trading Reference Links

When testing a system shouldnt all data before 4/10/2001 be ignore.
on 4/10/2001 was the date all exchanges converted to decimal.

nick

--- In amibroker@xxxxxxxxxxxxxxx, "tchan95014 <tchan95014@xxxx>" 
<tchan95014@xxxx> wrote:
> Hello,
> 
> 
> 
> 
> Here is an example on how I select a ticker to be included in a 
system 
> portfolio (a sure form of curve fitting, but...) I also touch upon 
how 
> I analyze the system base on one selected ticker trying to fine 
tune 
> this system. I hope to invite some discussions and debates, critics 
> and comments.
> 
> 
> 
> 
> 1) trade range: 1/1/99 - today
> 
> 
>    The range is selected to cover the big runup and big crash, 
hoping 
> to detect the system as well as target's response to the big swings.
> 
> 
> 2) A short term EOD system is used.
> 
> 
> 3) After optimize the system to death (another curve fitting), I 
> selected some parameters that look stable. (Thanks to Herman's 
> 3DCharts, I used it every time I do the optimization)
> 
> 
> 4) I started to run the system on a bunch of tickers, say N100.
> 
> 
> 5) I then select only those with, say > 20% RAR over the test 
period,
> 
> 
> (I would like to use MAR, but so far it is very difficult so I 
choose 
> RAR) I put those selected into a watchlist.
> 
> 
> 6) I then run tests on the watchlist.
> 
> 
>    a) I move the test range from 1999 - 2003 --> 2000 -> 2003, 
again I 
> chop off those not meeting a certain RAR, then 2001 -> 2003, then 
2002 
> -> 2003. I only want those that are profitable on every test and 
they 
> have to show to generate a certain RAR.
> 
> 
>    b) There will be only less than a handful candidates now.
> 
> 
> 7) I will then start to run the optimization again on the selected 
> candidates, just want to make sure the original optimized 
parameters 
> are still a good choice. If yes, good, a new portfolio is created, 
if 
> NOT, well, all bets are off, I might start again on the newly found 
> optimized parameters on the whole N100 again and start the whole 
> process to see if the new parameters are stable on SOME other 
tickers.
> 
> 
> 8) You might want to start testing the watchlist on any UN-tested 
data 
> range now. I did not.
> 
> 
> 8) It is a tedious process but interesting journey.
> 
> 
> 
> 
> OK, let us say I have the portfolio selected and system chosen. I 
will 
> go into each ticker selected to do some more analysis.
> 
> 
> As an example, I offer one below.
> 
> 
> 
> 
> Again, it is a short term EOD system, as you can see below.
> 
> 
> 
> 
> 	a) There are 240 trades from 1/1/1999 - 2/14/2003
> 
> 
> 		- a RAR% > 260% system
> 
> 
> 		- a MaxDD% > -86% system when position sizing = 100%
> 
> 
>             - I am eagerly waiting for the portfolio level position 
> 
> 
>               sizing feature to arrive.
> 
> 
> 	b) When the profit% per trade was analyzed
> 
> 
> 		- max. loss = -20.1%
> 
> 
> 	      - max. gain = +84%
> 
> 
> 		- average = 3.14% (all entries in the profit% per 
trade
> 
> 
>               column are averaged)
> 
> 
> 		- standard deviation (SIGMA) = 14.76%
> 
> 
> 		- average - 2 * sigma = -26.37%
> 
> 
> 			- this is larger than the max. loss, a good 
sign.
> 
> 
> 		- average + 2 * sigma = +32.66%
> 
> 
> 			- max. gain is much larger than this, also a 
good sign
> 
> 
> 			- FAT tail effect
> 
> 
> 		- It was found that if I align trades from max. loss 
to max 
> gain sequentially it takes the first 223 entries to get the gains 
to 
> offset the losses, it pretty much leaves only 18 profit trades to 
make 
> the extra profits taken home
> 
> 
> 		- 18 / 240 < 8% of the trades makes the day.
> 
> 
> 		- It also hints loudly that I have to take every 
trade, 
> because it is extremely difficult to be selective in choosing 
trades 
> and hit on those 8% big winning trades, they are so few and far in 
> between that I can not afford to miss any.
> 
> 
> 		- The FAT tail is very important, it means the losses 
have 
> been kept in check and profits have been left to run.
> 
> 
> 	d) Further analysis on the biggest losses
> 
> 
> 		- most of the biggest losses are the results of the 
GAPs, I do 
> not see any way to prevent these trades, I guess I have to live 
with 
> them.
> 
> 
> 		- Further check might be focused on the MAE (Maximum 
Adverse 
> excursion) on profit trades to see if there is any chance to use 
the 
> collected statistic to create a hard stop to further minimize the 
> losses.
> 
> 
> 
> 
> NOTE: this system has a very low profit factor of 1.34, but thanks 
to 
> Fred, it has also the following statistics on the last day of range
> 
> 
>    CAGR%  = 267.13
> 
> 
>    MDD%   = -68.89
> 
> 
>    MAR    = 3.88
> 
> 
>    UPI    = 56.35
> 
> 
>    KRatio = 3.30
> 
> 
>    MFlat  = 267
> 
> 
> 
> 
> Some more statistics on the analysis of profit % per trade:
> 
> 
>                     average  sigma    avg - 2 sigma    avg + 2 sigma
> 
> 
>    For ALL trades:  3.14%    14.76    -26.73           +32.66
> 
> 
>    Losing trades    6.14%    4.31     -14.77           +2.49
> 
> 
>    only
> 
> 
>    Winning trades   14.06%   15.33    -16.60           +44.73
> 
> 
>    only
> 
> 
> 
> 
> I can not offer any useful analysis on these statistics but I do 
hope
> 
> 
> TJ can offer All trades, losing trades, winning trades report.
> 
> 
> 
> 
> As you can see it is a very difficult system to trade. It is only 
an 
> example to illustrate the process.
> 
> 
> 
> 
> Hope you find these info interesting and tickle you brain.
> 
> 
> 
> 
> 
> 
> Thomas


Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx 
(Web page: http://groups.yahoo.com/group/amiquote/messages/)

Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/