Hi Paul -- I am a firm believer in performing statistical validation of trading systems. That means searching / optimizing over one set of data (the in-sample data), then testing over a set of data that has not been used to develop the system (the out-of-sample data). Trading systems that show good results over the out-of-sample data are more likely to trade well with real money. One of the things I have found is that trading systems that are selected on maximum CAR often do not trade well out-of-sample -- but you may have found systems that do.
Provided that your systems pass your validation tests, then you should be very glad that you have a lot of candidates. One way to handle this is to group the stocks you are willing to trade into watch lists containing perhaps ten stocks each. Run your system as a portfolio test once for each watch list and use the PositionScore feature of AmiBroker to determine the best individual stock to trade. There is a much more detailed explanation of testing and validation, including setting up portfolios and using PositionScore in my book, Quantitative Trading Systems. You can learn more about the book at its web site: http://www.quantitativetradingsystems.com/ Thanks, Howard On 8/4/07, polomorabe <[EMAIL PROTECTED]> wrote: > > Hello, > > I've been following the postings on this forum since I first began > using AB about two years ago, and thanks to the knowledge and sharing > mentality of the posters here, my knowledge of AFL has increased > greatly. I now have a question about trading system development. > > I've been getting experience with using the backtester, and I've > identified some promising trading systems, one with a CAR of over > 100%! During the backtest, I have limited the number of positions in > the portfolio to 7, to reduce overtrading. However, when I take the > same system from backtesting, and use it as an exploration, I get a > huge number (>200) of trading candidates at the start of each trading > day. > > My question is: without asking for details of specific trading > systems, how do other users of AB reduce the number of stock > candidates during the Exploration? During backtesting, I've found that > when I try to constrain the backtester to attempt to reduce the number > of candidates, it reduces the number of backtested trades but also the > percentage of winning trades. > > Maybe this is wishful thinking, but for me a winning system is one > that has a winning trades percentage > 50%, with a max system drawdown > > 15%, and generating about 10 stock candidates per day. > > I'd be interested to hear of other's experiences and views. > > Many thanks, > Paul > > >
