|
PureBytes Links
Trading Reference Links
|
> How do you deal with a string of losses? I'm not talking about filters.
> They are fine in testing but reality is another thing. At what point do
you
> say that the system has a problem.
I'll tell you what I do...
What you are developing your system, only test and optimize on a portion of
the data. This is the "in-sample" data.
After you have your complete system. Run the system on the out of sample
data. This is the first test of your system. If it loses money, scrap it.
If it made money after transaction costs (slippage & commission), then,
probably, decide to trade it.
Take the first 10 trades.
Do a t-test for population mean (might require some statistics study) to see
if your actual mean trade is less than your out of sample mean trade by a
significant amount. I use the 95% confidence level for significance
testing. If your actual trades are significantly less profitable than your
out of sample backtesting, scrap the system. If it passes (and it probably
will with so few trades), keep trading.
Continue doing the t-test periodically. Switch to a z-test when you have
more than 35 trades. If it ever fails, kill the system. Failing the test
means the market is different somehow than it was during your out of sample
period and the profitability of your system is in question.
Aaron Schindler
|