David,

You don’t talk abiut bootstrapping in your screencast. Is it considered historical?

Frank

Hi Frank,

I did under Wilmott in Quant-yes?

But yes, bootstrapping would be class of historical simulation as it “re-samples” (with replacement) historical data. The three-way VaR I stole (of course) from Jorion; Wilmott has three ways, too, he just serves it as simulation (forward versus backwards) versus parametric (delta normal).

VaR is funny, because, i can’t stress enough the metric is straightforward. K. Dowd says “it’s just a quantile.” To see this, IMO, helps explain why much of the VaR criticism is unjustified: the VaR metric is too humble (harmless) a tool to warrant such criticism. I think it helps to pause on the simplicity of the VaR idea, that we are selecting a point on a distribution tail ("want more confidence? okay, let’s scoot over into the tail more"). What’s complicated is how do we get the distribution, do we sample (historical), simulate (monte carlo) or use a formula (parametric). Getting the answer, if we use Jorion’s typology, has three approaches. Then each of those has a hybrid; e.g., hybrid parametric and HS in Allen. And so, the VaR metric is a merely a quantile (a point on a distribution), getting the distribution is the variety.
David