|
PureBytes Links
Trading Reference Links
|
Dear Jim,
I was using backtesting as a counter.
May be it sounds peculiar, but follow the lines:
I wanted to know how many
macd crossovers occured yesterday.
After the
c1=cross(macd(),signal());
I was creating an artificial buy-sell
buy=c1>0;
sell=buy;
then scan for 10/7/2001 and, here is the point,
hitting "backtest", I had at the left bottom of Results:
Number of trades:37 ,
ie I knew that 37 crossovers occured.
Besides theories, if you have any suggestion to answer
this 37, please let me know.
Now, I multiplied my possibilities by 1000 (or more).
"Composites" give answer to above question, for the whole
data, (not a single day), in less than a second.
For me, as you understand, it is great.
But, this new tool does not diminishes the value of the previous
one. No way.
I use the same method now, to test some doubtful composites
results. A gentle tool is always valuable.
Best Regards
Dimitris Tsokakis
--- In amibroker@xxxx, jvarn359@xxxx wrote:
> Jim:
> > I'd like to suggest that the user community establish a
> > standard that these systems are backtested to for the purpose
> > of including these results in the AFL library.
>
> TJ:> I wish other people on this group speak up about their
> > proposals so we can find something useful. Maybe a poll would
> > be useful?
>
> Are you 'sure' you want people to speak up? You might want to
> reconsider after I give my opinions :=) What I'm about to say
might
> be unpopular but I'll take a risk and speak anyway.
>
> I believe mechanical backtesting is "mostly" spurious and I don't
> think it is wise to rank AFL's by backtesting results.
>
> As proof consider well-known swing trader Gary B. Smith. He buys
> breakouts over trendlines. I've modeled his simple system and it
> generates poor returns.... but Mr. Smith's trading has made him a
> millionaire. I'm no millionaire but I've turned in some good
trades
> lately using simple divergences. These too fail under backtesting
> but the profits in my account are no less real.
>
> Here's why backtesting underperforms a good trader: stock
selection.
> I think Smith and I use the same approach. I use Explorations to
> generate a short list of candidates to trade. I put the results
into
> a watch list, select the first stock, then use the down arrow to
flip
> thru the charts. If I see a price and volume pattern I "like" then
I
> buy. Patterns I "like" are based on accumulated experience from
> numerous trades over the years. When I use AmiBroker I never
> backtest... I have no need or desire to have little white and red
> arrows cluttering up my charts!
>
> The problem with backtesting is that it buys every candidate, even
> the ones with bad charts. Traders outperform the computer because
> the eye and brain are very good at visual pattern recognition and
> integrating other factors such as how the major indexes "feel",
> economic events, sector rotation, etc. Compounding the problem is
> that pattern recognition and external data integration are updated
> daily by my brain, whereas the mechanical system holds the stock
> until a rigid exit is triggered.
>
> If one is serious about mechanical backtesting then there is no
need
> for a charting program. Just use a database or spreadsheet with
the
> trading algorithms (see the Mechanical Investing threads on Motley
> Fool). Let the database or spreadsheet to tell you when to buy and
> sell. No need to ever look at a chart again.
>
> To sum up: backtesting tests the *computer* as trader not the
*human*
> as trader.
>
> Ranking AFL's by backtesting scores can mislead people. A negative
> return system could generate positive returns in the hands of a
good
> trader through careful chart selection. Conversely, a system that
is
> labelled the "best" may do poorly in the hands of some traders.
>
> Now that I've said that, I don't mind if AmiBroker has backtesting
if
> most people want that feature. Good programs come with a full
chest
> of toys and its up to each user to decide which toys to play with.
> But I hope that the focus of AmiBroker stays primarily with
charting.
>
> Regards,
>
> Jim Varney
|