|
PureBytes Links
Trading Reference Links
|
> Currently I normalise the data series or the close value, then I put a linear
> line on that data, then I take the differences for StdDev calc.
here is an example of why not to pre-normalize your data before finding the
regression while I am waiting for my morning coffee -
Assuming same time range for stock A and B
stock A - start at 10 and move to 11.
stock B - start at 20 and move to 40.
If you pre-normalize, both end up moving say from 0 to 1!
That is called information loss caused by transformation.
Since you are interested in the slope, etc. information of the
regression line, but the transformation you want to apply
will modify that very information, you should do that through
some other means.
>
> Are you suggesting that I do this : Take Close (not adjusted), place linear
> line on this data, take the difference btw line and closes at 5 min data
> points, THEN normalise the DIFF VALUES ??
after the example about, here is one of the proper ways to standardize your
deviation data - use percentage of the final regression value.
say you get for the same example,
stock a - regression line slope 0.2, deviation $2, final regression value $11.5
stock b - regression line slope 1, deviation $2, final regression value $42
then normalize the deviation from dollar value to percent,
stock a - 2 / 11.5 x 100
stock b - 2 / 42 x 100
as the SD are now in percent, you can compare them.
>
>
> My scanner works great : 400 stocks of 5 min data in 50 secs...I maintain
> ASCII files on my hardrive care of QCollector.
as you mentioned that you are working on 1 day data only - that is sort of slow :)
Lawrence
>
> Rgds
>
> Ian McVicar
>
=====
Lawrence Chan http://www.tickquest.com
Transform market data into opportunities
|