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

RE: A question about system design.



PureBytes Links

Trading Reference Links

At 07:09 PM 7/15/00 -0500, M. Simms wrote:
>What you are saying then is that backtesting large samples is fruitless ???


Of course its fruitless! If it wasn't, we would all be wealthy by 
now. Just make a system or buy one which has a nice backtest over 
300 trades and retire in a few years. Its so easy right?  :)

But really, I'm saying your idea of large is probably too small to
give *me* much confidence. How many trades did the Demark system have
before it blew up - enough to give you some confidence in it? As I 
said, I have seen very little or no correlation between sample size 
and future results in systems we backtested and then traded which 
had as many as several hundred trades. That whole porfolio of
systems made money, but only a few systems in the entire group 
carried the load. Maybe you have seen otherwise in your actual 
testing/trading. I've seen tons of systems quickly break down with 
200 trades and nice equity curves. We traded a system once which 
had no losers in 98 trades in the backtest, but it broke down 
immediately in realtime trading!
On the other hand, the stock system I mentioned with thousands of 
trades I do have some confidence in. I like seeing 10,000+ trades
do well, and testing 75 stocks and seeing every single one make
money over the test period. Now that inspires at least a little
confidence that maybe this thing might actually work going 
forward. I just don't get that same feeling from 30-300 trades.

regards,
rich