Normalizing Long and Short Weights in Eigenportfolios

Summary

The document asks how to normalize eigenportfolio weights when some components are negative, in a setting based on statistical arbitrage in U.S. equities. It contrasts dividing each weight by the signed sum with dividing by the sum of absolute weights. The response suggests that the appropriate choice depends on the intended interpretation: signed weights represent long and short positions, while absolute weights can describe exposure magnitude.

The answer is brief and explicitly framed as intuition rather than a reading of the cited paper. It also suggests using the number of weights as a denominator, but does not derive this recommendation or reconcile it with either normalization method. No portfolio example, return calculation, or evidence is supplied. The key practical point is that normalization should follow the portfolio constraint or quantity being measured; the document does not establish one universally correct formula, and its response should not be treated as a complete construction procedure.

Key ideas

• Negative eigenportfolio weights represent short positions alongside long positions.
• Signed-sum normalization and absolute-weight normalization imply different interpretations of the portfolio.
• The appropriate denominator depends on whether the goal is to normalize net value, returns, or exposure magnitude.
• The response is tentative and does not derive or validate a general normalization rule.

Tags

# When you have negative weights in the context of portfolio construction, what is the correct way normalize them?


# When you have negative weights in the context of portfolio construction, what is the correct way normalize them?












For context, I am building an eigenportfolio following the conventions of Avellaneda and Lee Statistical Arbitrage in the U.S. Equities Market (2008), and I get negative weights for eigenportfolios 2,3,.., and so on.

I wanted to know if the correct way to normalize is to divide each element by the sum of elements, as you would do when you have positive weights? I was thinking of taking the sum of the absolute value and then dividing each element by that.

I was having some difficult time visualizing implications under both cases, and so was unsure which method is more appropriate...

I would be grateful if you could provide some kind guidance on this matter.

## Answer by Hasselhoff (score 0)

https://quant.stackexchange.com/a/80333

Didn't read the paper, but my intuition says depends if you need to normalize for the purpose of returns or value. if the weight is negative, that just means it's a short position, so for value its a credit to the account when you enter, while the positive weights are a debit, and you want to leave them as is. But the for returns, as long as you have the sign of the return wrt to gains or loss, then you could use absolute value of weights. Either way, you use n = count of weights as your denominator.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.