PureBytes Links
Trading Reference Links
|
--- In amibroker@xxxx, farriners@xxxx wrote:
> Hi,
> As a newcomer just getting into optimizing, it would be really
> helpful to have a rough idea what more experienced guys expect from
a
> reasonable system. I am not asking for trading secrets/formulas,
just
> an idea on returns to aim for to help me and other newbies.
> info' such as:
>
> period covered
> % return
> number wins/losses
>
> (also if possible $invested per trade, broker fees)
>
> ie if I am getting 15% on 30% wins/losses is this above average,
> average or "dont give up the day job"
Ill tackle a few of these questions. First there is no right or
wrong...just what you as an individual feel comfortable with...
Period: does not need to be very long if a variety of market
conditions are included. Right now it is easy, look at AMAT for
three years. There was a moderate upslope in the beginning, the wild
bubble in late '99 early '00, the selloff in late '00 early '01, and
the choppy sideways action in mid '01. That is a good sampling of
different market conditions. You don't need to go back many years.
% return will be all over the map depending on the security, period,
and the amount of over optimization (curve fitting) you used. Look
at a couple of major market indexes. Can you beat them in sideways
and down markets? Is your return higher and your drawdowns lower?
Remember that it is very difficult to beat buy and hold durning an
extended bull run like late '99.
The number of wins and losses, or hit rate, if you will, is very
subjective. Almost any trading idea will be right about one third of
the time, no matter how poorly it is constructed. Put another way,
throwing darts at the financial page will give you about 35% winners.
Some people, and many beginners, want to see very high hit rates on
the order of 70-85%. Most systems with very high hit rates like that
have a very small return per trade and small returns overall. Some
of the very best systems have low hit rates in the 30-40% range but
catch the occasional huge winner for overall good profitability.
I prefer hit rates rates above 50% if possible. I can't stand a long
string of losses with no winners. Some people can. You need to
figure out your own comfort level in this area.
Some other things to look at are:
Average winning trade, this should be a couple hundred dollars or
more, to cover commissions, slippage, inability to get the trade
executed promptly, inopportune phone calls, etc. Also recognise in
real trading you will probably only make half the profit that the
optimizer made.
$ Win/loss ratio: This will be low to less than 1 for high hit rate
systems, and must be 3 or much better for low hit rate systems. If
you are going to lose money on two out of three trades they better be
small losses compared to the wins.
Maximum drawdowns should be better than the drawdowns from buy and
hold, otherwise your system is rather ineffective. My philosophy
is, we trade to reduce losses, the winners will take care of them
selves. If your losses are no better than buy and hold dips, why
bother trading.
When you get serious about a trading system, examine the trade by
trade display on the chart. Is the system buying and selling at
reasonable places. Did it totally blow it somewhere and be on the
wrong side of your desired position for a period of time. Those kind
of trading errors need to be fixed by modifying the system to avoid
or minimize the problem areas.
Many times the most profitable optimization is not the preferred
one. The optimizer will "get lucky" and catch some big move and
screw up the rest of the trades. This is why you need to examine the
other statistics and the actual trades on the chart. Look at the top
15-20 results and see how the other factors vary. Plot several of
the results on the chart. I find that that if I get the same
parameters for best profit factor and best ROA as for max profit
then I may have a reasonable optimization run.
Well, by now you should be in data overload, so I will end it here.
Needless to say there is a lot more to optimizing backtests than just
doing a couple of quick runs.
Cheers
Trader
|