[R-sig-finance] bayesian signal classifier

Mon Nov 28 02:21:14 CET 2005

```Assume you have two decisions to make buy or sell. Then if you assume
the classifications errors have the same cost (this  can be
modified suitably for cost) then a naive bayes classifier that minimizes
the error rate is simply

prior * likelihood

So assuming that you have a prior computed in-sample from the timeseries
you compute for each date the posterior decision buy or sell based
on the returns+ the signals (i.e. the feature vector).

Date            Returns        signal 1           signal 2            ...
---------------------------------------------------------
01/01/1990
.
.

I am sure there are folks on this list who can throw more light on this
R has a lot of classification related machinery that can be put to use
that will do better than this Bayes classifier. You may want to refer to
these two books.

http://www.amazon.com/exec/obidos/ASIN/0122698517/kriskumar-20/104-8074544-7680720
http://www.amazon.com/exec/obidos/ASIN/0471056693/kriskumar-20/104-8074544-7680720

And finally the V&R yellow book describes some of the classifiers that
are available in MASS.

Kris

paul sorenson wrote:

>The issue I am curious about is how to classify various signals (eg
>price, volume, MACD, etc) into to buy, sell, or hold?
>
>Assuming I could "tokenize" various attributes of signals (value, 1st,
>2nd and 3rd derivatives, crossing, etc), would it be feasible to take
>these as inputs to a (trained) classifier which then outputs some number
>between 0 and 1 representing buy, hold, sell?  The analogy I am thinking
>of is a Bayesian spam classifier.
>
>My background is in engineering and I have only basic statistics
>knowledge.  I have been using R for a couple of years now mostly for
>graphic output.  I have a reasonable grasp of the language but I'm not
>strong on the underlying theory of the statistical functions.
>
>R has a number of packages which deal with Bayesian statistics but I
>don't have the knowledge to join the dots from there to a classifier.
>
>Any pointers would be most welcome.
>
>cheers
>
>_______________________________________________
>R-sig-finance at stat.math.ethz.ch mailing list
>https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>
>
>

```