Yours is a single-asset absolute VaR (absolute: includes the drift, µ) but the derivation is assumes multi-asset (ie, multiple weights) relative VaR, so per 7.17 below, the marginal VaR just multiplying (scaling) the portfolio's marginal volatility, ∂σ(P)/∂w(i), by the deviation, α; e.g., α = 1.65 if 95% confident. I hope that's helpful,
Hi @sapozan Yes, that's observant! It is excluded. Your formula is not in matrix format, but could represent a multi-asset portfolio such that we can think of the generalized (multi-asset version of your) Var = µ + σ ∗ z. The proof is not using this (which includes the drift and is called absolute VaR) for portfolio VaR; rather the formula is using the multi-asset version of this: Var = σ ∗ z; i.e, you are correct the drift is omitted. Because this marginal VaR is "keeping it simple" by only looking for the marginal volatility where portfolio variance is the super-convenient w^T*Σ*w where Σ is the covariance matrix and w^T is the transposed weight vectors. This does omit the drift and captures the relative VaR: the VaR relative to expected future value such that drift is omitted. So it's just for analytical simplicity, there is an analogous absolute VaR although I can't recall seeing it analytically. My hunch is that it's just easier to simulate given the drift is linear and doesn't fit in nicely.
So basically the justification is: we call it relative VaR (and keep things very simple)! If we want to include the drift, it's an absolute VaR and it's a bit more work. Thanks,
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.