|
PureBytes Links
Trading Reference Links
|
Mike,
Boy this is a very subjective subject. I have generated
systems that generate over 80% profitable, but I don't trade
them, as the average trade size is too small and the trade
count too high for my liking. If you like scapling and if
you can autotrade them without slippage, maybe this is to your
liking? These kind of system can also generate Sharpe
ratios greater than 8.0.
Typically, I look for 3 to 4 sharpe ratio levels given that that
the system generates "sufficient' number of trades/period to
insure that the Sharpe ratio is not a fluke. These systems
typical generate 45% to 60% winners with a Profit Factor
greater than 2.5. I become suspicious on the validity on
of a system test that generates Profit Factors greater than 4
unless it's backtested with tremendous amount of data.
I do not concentrate on the percent winners. I let Sharpe
drive the strategy development. About the only time
Sharpe ratio will fool you is when there is an unsufficient
number of trades in the test per period and per the entire
data test set. You learn be suspicious of high Sharpes too.
Here are some observations:
a) The more the system trades, the smaller the avg trade,
the higher Sharpe ratio and the higher the percent profitable.
So if this is true, then I look for a compromise, on these
values. To do this, I export backtest results and conduct
a 3D study to insure that I pick a good set of variables and
keep me from picking an isolated fluke result set.
b) The simpler the system, the higher the chances that you have
not curve fitted.
c) The more data I test with the better I feel about the
validity of the system.
d) The smaller the avg trade size and
typically the higher number of trades, the more
sensitive you need to be on the slippage and whether your
simulation uses stop, limits or mkt orders.
e) Be suspicious on your findings if Sharpes are greater than 4
or the ProfitFactor is greater than 4 or the percent profitable is
greater then 65%.
f) I don't have the exact ratios but, be leery of too few
trades per a period of time. The changes of a random fit
are just too high. For example a system that trades 100
times over the course of 1 year is simple too risky for me
to depend on any of the result numbers. This kind of
result needs to be verified across many years of intraday
data.
Regards,
Ernie
ebonugli@xxxxxxxx
|