VAR

Discussion in 'David's Notebook' started by amir2403, Jul 9, 2012.

  1. amir2403

    amir2403 New Member

    Hi,

    I need to unferstand:

    1) the impact of correlation in computing VAR through MC method ?

    2) under what conditions / assumptions we can compute VAR through MC without volatilities and correlations and how exactly (for multi asset portfolio)

    Appreciate help on this...
  2. Aleksander Hansen

    Aleksander Hansen Well-Known Member

    1) when simulating the trajectories, the paths are not independent of other assets' path. See Cholesky decomposition for the normal case; and see Jorions quantitative chapter, or google for specifics.

    2) you can't
  3. Just to add to Alek's (1) from an FRM perspective: we tend to refer to two types of correlations (keep in mind "correlation" is linear dependence, so it's pretty narrow) in a portfolio, over time (aka, auto or serial correlation) and across the assets (what i like to think of as cross-sectional). Realistically, both tend to occur.

    Among the three major VaR approaches:
    • The drawback of naive parametric VaR approaches is that its cross-sectional correlation tends to be simplistic and, under square root rule scale, it totally omits serial correlation; e.g., even in Basel, scaling 1-day VaR to 10-days with *SQRT(10) assumes i.i.d.
    • Boostrapped (historical simulation) VaR improves by incorporating a realistic (historical as they actually occurred) cross-sectional correlations but its simple variation omit serial correlation, albeit with chances for improvement in its variations (see Dowd's non parametrics); but finally
    • One way to view the superiority of MCS is that it rather easily incorporates both correlation types, including superior "correlation over time " (path dependency) and correlation between assets (with Cholesky, as Aleks says)
  4. caramel

    caramel Member

    Hi David

    I cannot find any practice questions on Linda Allens Putting Var to work . Are there any thanks
  5. Aleksander Hansen

    Aleksander Hansen Well-Known Member

    Don't take my word for it, but I don't believe there is... several other PQs that cover the same material though. Besides, the Linda Allen reading should be retired IMO.
    • Like Like x 1
  6. Hi caramel - Please see http://www.bionicturtle.com/forum/t...-allen-chapter-3-putting-var-to-work.5250/for further justification, but it's true that we skip some chapters, chief among them are Linda Allen's. Although the current semester's sequence (i.e., the period up to the November exam), which is currently in T3, will definitely "go deep" with value at risk (VaR) questions. So, in addition to the existing PQs, there will be more coming. Thanks!

Share This Page

loading...