Can I do some like [factor neutralization]

General
1

Can I do some like [factor neutralization]

2

Your question is not understandable.
What are you trying to accomplish (example)?

3

In my counry, when to use one factor. we use method in data manipulation.
we will Standarize,Neutralize and Winsorize the factor

1 Like
4

I am not sure that this is a complete answer but you might look at zscore() to start (Learn & Connect > Factor Reference).—Jim

5

zscore is Standarize， the most i want is Neutralize

1 Like
6

I don’t really understand the question either, but here’s how to neutralize a factor under a certain condition. Let’s say your factor is Pr2SalesTTM and your condition is that IWN is below its 200-day SMA. Then you could write Eval(Close(0,GetSeries(“IWN”))<SMA(200,0,GetSeries(“IWN”)),NA,Pr2SalesTTM).

For Winsorizing, you can use the Max and Min functions. For example, if you want to maximize Pr2SalesTTM at 30, you would use Min(Pr2SalesTTM,30). Or, in place of 30, you could use a multiple of the standard deviation of a cross-section of the function using the ZScore function.

7

Are you talking about something like a risk model? Where you determine your exposure to common factors and then remove stocks which are highly correlated to that factor or adding in stocks which have negative correlation to it?

I can see the application of this if, for example, you had a strategy which was something like R&D to marketcap. And maybe it picked up lots of value stocks along the way and maybe you don’t want this.

I don’t know of any tools that we currently have which will help this. If it was a specific case and I didn’t want value or high yield or whatever, I could also just make a rule excluding the top 5 - 10% of stocks with the highest undesired factor rank.

Is this in the vein of what you are asking?

8

I think what he was referring to is neutralization in the sense of removing single factor exposure to other factors or a ranking system.

9

My take on some of what Azouz links to. It is pretty complex: no doubt about it. I do not do neutralization, for example, but I might take a lesion from it.

Azouz is a highly competent programmer. I am less aware of his mathematical degrees but I can say he picked up an advanced understanding of mathematics somewhere based on my relatively brief interactions with him.

Neutralization seeks to find interactions if I understood correctly.

What could be better for finding interactions than neural-nets when P123 makes AI/machine learning available?

But here is the other thing that is a big plus for P123. If you step back from the complexities and jargon this is mostly just linear algebra taught in high school with some continuation of matrix multiplication taught in lower-division college math courses. Admittedly for those majoring in math or physical sciences.

But it is not what we often think of when we think of AI (although I often have trouble seeing the line separating them).

So, this is my point: often linear algebra results can be plugged directly into a ranking system.

TL;DR: P123 is the place to be and the advanced capabilities that P123 enables are not fully recognized by most of P123’s members.

I suspect Azouz fully recognizes this and more. As would the other members of Numari from a marketing perspective, I think.

10

There is no way of neutralizing a factor in P123, not in the risk model sense.

Neutralizing a factor means a portfolio has no net long or net short exposure to the factor. For example, if you took the average ranking of a factor across a portfolio of 50 stocks, you would get 50, aka the portfolio has a median ranking of that factor.

1 Like