Forum Archive Index - April 2004

[Phaedrus <Phaedrus@techemail.com>]

T100,
     I am a great believer in optimisation of TA parameters. Not everybody 
approves of it and you must be aware of the dangers it can pose. Taken to 
extreme, it can lead to curve-fitting. One way to guard against this is to 
split your data into 2 lots. Perform all backtesting and optimisation on the 
first set, then apply the optimised indicators to the second set of data. Judge 
the efficacy of the optimised indicator on these results, rather than how good 
the initial optimised results were. OmniTrader does this automatically - 
backtested optimised parameters are then forward tested on later data and the 2 
sets of results compared. If you get spectacularly good backtest results and 
very poor forward test results, you have been curve-fitting. There is another 
clue. Say you find that an RSI period of 20 is the optimum for a particular 
stock. Look at how well other values performed. If the second best was 43, the 
third best 8 and the fourth best 30, clearly the 20 result is a fluke and you 
cannot expect it to be robust, working well in the future. However if the 
optimum period was 40, second best 39, third best 41, fourth best 38, the 
optimisation is valid and figures of around this value really do work best for 
that particular stock.
 To answer your specific questions :-
(1) Not sure whether you mean indicator periods or backtest periods. There are 
no "appropriate" indicator periods to backtest. Use a range from way too low to 
wildly high - the computer will find the optimum value for the period tested. 
For the backtest period, longer is better. A bare minimum might be a couple of 
years, but 10 or so would give more meaningful results.
(2) MetaStock backtests the periods loaded. This is set under "Load Options", 
or else you can set it so you are prompted for the date range you wish to 
backtest and optimise on when each chart is opened.
(3) Use the "Compare" function, list all the indicators that you are interested 
in, and in one hit you backtest, optimise and compare the results of them all, 
with results listed in the order of percentage gain - best indicators first. 
This would be a good initial step when you start to design a system for a 
specific stock.
(4) There is no point doubling up on indicators simply for the sake of it. For 
example RSI and RVI are very similar, so there is not much sense in running 
both. One approach you may like to consider is using an oscillator, a trend 
indicator and a volume indicator. With trend indicators there is more point in 
doubling up because methods can vary widely, from trendlines to moving averages 
to indicators such as DMI, etc.
 Keep things as simple as possible. You will end up with your own favourite 
indicators just as others do. There is no reason at all why you should use the 
same selection on every stock though. Use what works best - all stocks are NOT 
the same.
 One last point. Make sure that when you examine the backtest results, you look 
at the "Most consecutive losses" figure. I have had quite profitable systems 
that I personally have found unuseable - I lose confidence in any system after 
about 3 consecutive losses, no matter how good it may prove to be overall.
 You do realise that you can optimise for many different things? Overall gains, 
best hit rate, best win/loss ratio, least consecutive losses, lowest drawdown, 
etc etc.

    Regards,
               Phaedrus.

_____________________________________________________________
Are you a Techie? Get Your Free Tech Email Address Now! Visit 
http://www.TechEmail.com

----------------------------------------------------------------------------
To remove yourself from this list, please use the form at
http://www.sharechat.co.nz/chat/forum/