|
PureBytes Links
Trading Reference Links
|
Excessive curve fitting usually means that you have fitted more coefficients
than you have data points. This is obviously over fitted.
Usually curve fitting is when the price is traced by an equation. Usually
some compound sine series. It is possible to build these and use the
resultant equation as a proxy for price. The trouble is people then let the
equation forecast ahead of time. Any proxy for price has little predictive
power (if any). I was going to suggest beyond a few days but saying even
that is probably too much.
OTOH I don't think there is anything wrong with specialised curves for
specific stocks. Within stocks its not a surprise to have seasonality
perhaps (for example) except with one caveat. Check the robustness of the
fitted indicator by moving periods etc around (i.e bigger smaller) to see
how sensitive it is. Im suggesting in the case of sine series change
periods. Is there a harmonic? If its sensitve its probably an
inappropriate curve fit. Preferably when (not if) the stock changes
characteristic you want robustness where you'll fail gracefully and
hopefully realise in time before you've completely shot your wad.
A sensitive indicator OTOH is not curve fitting.
Multiple, different functionality, confirmatory indicators is not curve
fitting. EG when volume upticks and price heads up as shown by MACD ROC is
not curve fitting.
Can you seee the difference. Curve fitting is realm of time series analysis
and Box Jenkins, AMIRA, ARCH and GARCH. IE forecasting.
Credo #3 DO NOT EVER PREDICT. Anticipate and react. Let the market pull you
in and throw you out by careful use of stops. (NB Exit is more important).
P
----- Original Message -----
From: "David Holzgrefe" <dtholz@xxxx>
To: <amibroker@xxxxxxxxxxxxxxx>
Sent: Monday, June 04, 2001 11:09 PM
Subject: Re: [amibroker] Steady Movers
> Thanks P
>
> Everyone of your credo's is part of my trading rules :)
>
> I was just trying to work out what curve fitting really is
> I think its where you develop a system around every move of a stocks prior
> performance .
>
> What im trying to understand is to what level do you have a system that
> suits a certain stock or type of stock ?
>
> Here's a simple example XXX stock has backtested to perform well with the
> mov 10 , 30 cross but shows poor results over a portfolio test ,
> Do I forget it and look for something that suits a wider range of stocks
> or keep it just for that stock , if I do is this regarded as curve fitting
?
>
> That was a simple example it would mostlikely be several indicators that
> when combined show good results .
>
> I should know the answer having read may times about it but it just wont
> sink in :))
>
> Thanks David
>
>
>
>
> ----- Original Message -----
> From: "server not recognized" <winchp@xxxx>
> To: <amibroker@xxxxxxxxxxxxxxx>
> Sent: Monday, June 04, 2001 10:25 PM
> Subject: Re: [amibroker] Steady Movers
>
>
> > FWIW.
> >
> > DO NOT curve fit. It lulls you into a false sense of security.
> >
> > My creedo#1 is "Hope is the enemy of greed".
> >
> > Therefore a certain skitishness , like a nervous cat, is necessary to
leap
> > out the way the MINUTE you think its danger. It doesn't matter that you
> do
> > ten trades, all wrong, if you only loose 5% per trade. But when you're
> in,
> > if after all that time for contemplation, you get the direction wrong -
> GET
> > OUT.
> >
> > DO NOT constantly seek the 3000% trade. Be happy to take 10%-20% per
> trade
> > 10 time a year. Creedo #2 "Compounded gains are my best friend".
> >
> > DO NOT over trade.
> >
> > DO NOT concern yourself with "who has got the cheapest brokerage". At
say
> > $30 /trade for up to $10,000 its less than worth worrying about. You'll
> > lose more on hanging on an extra day in a losing trade than you'll ever
> save
> > on commissions. Guaranteed.
> >
> > DO NOT be afraid to pull the trigger on a winning trade. As the stock
> rises
> > put stops closer to the price. If you don't have an automatic sell at a
> > stop RING YOUR BROKER, or INTERNET SELL. If the price drops back don't
> > start hoping that it should go back. Do not break credo #1.
> >
> > DO NOT LOOK Back. If you sold stock and after it goes back into the sky
> > don't bitch. Buy it back if you think it's really going somewhere -
else
> > STOP BITCHING.
> >
> > DO NOT trade your account to nothing. If you've made say 50%, take 20%
> off
> > the table.
> >
> > Thats the start of my rule set. Just remember some organisations have
more
> > people with more intense algoritms, faster computers and more man hours
> than
> > you. Amibroker is auseful tool but the above rules will probably make
you
> > more money.
> >
> > DO NOT BE UNHAPPY! Figure out where your sleepness night stress level
is
> ,
> > back off a bit, sleep well and keep perspective.
> >
> > P
> >
> > ----- Original Message -----
> > From: "David Holzgrefe" <dtholz@xxxx>
> > To: <amibroker@xxxxxxxxxxxxxxx>
> > Sent: Monday, June 04, 2001 8:00 AM
> > Subject: Re: [amibroker] Steady Movers
> >
> >
> > > Trader10
> > >
> > > Thanks for the insight into your trading methods I can see the you
will
> > > appreciate the weekly scans that will be available in ver. 3.7
> > >
> > > You may find the piece of code bang for bucks that Geoff M used in
his
> > > searched handy for ranking too .
> > >
> > > Can I ask do you back test your results ?
> > >
> > > If so what criteria do you regard as successful result ?
> > >
> > > % return
> > > % win
> > > a comparison to buy/hold
> > >
> > > I must admit I've been using T/A software for over a year now and
still
> > > haven't come to full understand backtesting. but I can see it as a
> > parallel
> > > to studies I did on statistical quality control many years ago.
> > >
> > > I guess the new addition to the family may help me understand it more
(M
> > > Prings breaking the black box ) .
> > >
> > > whilst I'm flapping my gums (fingers in this case) another topic that
> I'm
> > > still trying to understand curve filling , A hot one I know I still
have
> > the
> > > bruises
> > > from another forum on that one :))
> > >
> > > Its regarded as a no no but ? is there value in developing a set of
> rules
> > > for an individual stock .. ?
> > >
> > > Ok I have raised a few trader type questions here that I,m interested
in
> > I'd
> > > love to see a 1000 word essay on each of the topics from you all *W*
> > >
> > >
> > > LOL
> > >
> > > Regards David
> > >
> > >
> > >
> > >
> > >
> > > ----- Original Message -----
> > > From: <traders10@xxxx>
> > > To: <amibroker@xxxxxxxxxxxxxxx>
> > > Sent: Monday, June 04, 2001 4:40 AM
> > > Subject: [amibroker] Steady Movers
> > >
> > >
> > > > My current project is to identify strong, steady trending stocks.
> > > > Over the last couple years I found that while I was busy chasing the
> > > > latest momentum stock du jour I was missing some of the longer term
> > > > really nice trending stocks.
> > > >
> > > > I picked 100 or so issues that have had strong runs lasting several
> > > > months or more. These stocks may have strong enough ROC to bubble
to
> > > > to the top of a momentum scan, but typically they do not sort to the
> > > > top 10-20 positions on a ROC type scan. Yet there are periods in
the
> > > > last few years that I would have really liked to be in them.
> > > >
> > > > The runs last long enough that picking them up after a month or two
> > > > is acceptable.
> > > >
> > > > My initial work has been to find indicators or patterns that do
catch
> > > > the trending period on those 100 stocks.
> > > >
> > > > Once a scan can catch the runs then I check to see if
> > > > that "indicator" will have so many false entries as to drive me
> > > > crazy with excessive trading, commissions and tied up capital.
> > > >
> > > > If the trading idea still holds up, I then check it against the
> > > > universe of stocks that I normally follow. That is ~2500 stocks
> > > > prescreened in TC2000 or QP2 for price > $4, avg vol > 300k or so.
> > > >
> > > > AmiBroker is very handy for this use of backtesting as a screening
> > > > process for trading ideas. I can whip up a few lines of code in AFL
> > > > and quickly backtest against the 100 stock portfolio to see if it
has
> > > > any merit for this type of trade.
> > > >
> > > > I can tell you that in many cases doing this on a weekly basis is
> > > > superior to trying to make it work on a daily basis. Fortunately
> > > > there is no great hurry to enter the trade. Once this type of scan
> > > > has identified the initial conditions the candidates can be put into
> > > > a watch list and further evaluation can be done on the concentrated
> > > > list.
> > > >
> > > > Things that need to be examined at that point are possible prior
> > > > basing, current earnings situation, smoothness of trend and applying
> > > > the oscillator or pattern of choice to enter the trade.
> > > >
> > > > Is it working yet...no, always looking for ideas.... I have
> > > > evaluated a lot of ideas so far and naturally no single indicator or
> > > > pattern does the job...what else is new :-)
> > > >
> > > > Preliminary evidence points to a voting scheme with 3-5 inputs for
> > > > entries and maybe 1 of 3 or 2 of 3 conditions for exits.
> > > >
> > > > Looking for some code for indicator divergence as one of those
exits.
> > > >
> > > > Well, enough for now. This gives some idea of my current project.
> > > > Trader
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > Your use of Yahoo! Groups is subject to
> > http://docs.yahoo.com/info/terms/
> > > >
> > > >
> > >
> > >
> > >
> > >
> > > Your use of Yahoo! Groups is subject to
> http://docs.yahoo.com/info/terms/
> > >
> > >
> >
> >
> >
> >
> > Your use of Yahoo! Groups is subject to
http://docs.yahoo.com/info/terms/
> >
> >
>
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
>
>
>
|