I have a piece of daily volume and a piece of daily trades data, now I divide them and I get volume/trades as a factor. This factor has positive value and I want it to be normalized to predict the next-day return by its sign. My solution is to revise the factor as
factor = factor - factor.rolling.mean(window). In this way the factor indeed contains both negative and positive value. However, it uses window as an extra parameter. The problem is: Is there any normalization method without using an extra parameter? Thanks in advance!
$\begingroup$
$\endgroup$
5
-
$\begingroup$ What about standard way: (factor(t)-mean)/stdDev ? $\endgroup$– K. RomanCommented Jul 25, 2023 at 6:25
-
$\begingroup$ That is known as standardization, while the OP is asking about normalization (but they use the term incorrectly, so perhaps your solution is fine for them). $\endgroup$– Richard HardyCommented Jul 25, 2023 at 7:17
-
$\begingroup$ Actually I only need the sign of the factor, so this standard method is equivalent to factor minus mean. Yet the mean needs a window. $\endgroup$– atlantic0ceanCommented Jul 25, 2023 at 7:17
-
$\begingroup$ @RichardHardy Here I didn't distinguish standardization and normalization. What I request is transforming the original data into a piece of data distributed equally into negative side and positive side. $\endgroup$– atlantic0ceanCommented Jul 25, 2023 at 7:20
-
$\begingroup$ @atlantic0cean, OK, so what you want is neither std. nor norm. but rather some other transformation (perhaps mean-centering or median-centering). It is usually helpful to stick to established terminology, so it could make sense to edit the post accordingly. $\endgroup$– Richard HardyCommented Jul 25, 2023 at 8:57
Add a comment
|