If alfa=90% and t=20 the L is at 18. We take the 3rd largest loss as per the core reading example. Why are we not taking the average of the top 2? Or the second largest?
The definition of VaR is effectively the largest loss before you get into the tail. So 20 losses @ 90% = the 18th loss. If you take the second largest this is the alpha=95% VaR. Averaging would give you a tail VaR.
Another quick question relating to calculations of Value at Risk. I'm on Q&A part 2, solution 2.18. The formula given for VaR = inverse normal distribution * VOLATILITY * HOLDING. Is this formula an alternative solution every time the losses are normally distributed? I.e instead of using miu + std. dev * inverse normal dist, we are forced to use the above formula just because the data given in the problem relates to volatility and holding instead of average loss? Would this formula apply to any other statistical distribution? On the exam should we expect VaR only based on normally distributed losses?
And last question on question 2.18 part iii. How is the solution derived? Which formula is used here? miu + std dev * inverse of normal or volatility * holding * inverse normal?
Another question on VAR calculations. The ActEd Q&A materials are using interpolation for the exact Z value for 99% probability. Is this something we are expected to produce in the exam? I have not gone through all the papers yet to verify the level of accuracy expected. The interpolation takes an extra time....
When you get a chance to look at past papers, you will see the examiners often accept a range of approaches to estimate these values.
Thanks Simon. I will look and see. Hopefully interpolation is not necessary. Would you also have time to answer the other questions above?
The formula is the same - volatility is the relevant measure of s.d. here. mu is assumed to be zero (unless another assumption is better!). As "mu + std. dev * inverse normal dist" is "per unit" we need to multiply by the size of the holding. The basic idea is the same, yes. See Q&A 2.17 - you would simply be starting with a different distribution. Not necessarily, but if you are expected to do calculations, one might reasonably expect the relevant significance points to be given or easily looked up in the Tables.