If you have long and shorts from different countries in different sectors, would VaR still be a good measure of your risk?

I think it is as good a risk measure as long and shorts from the same sector. But is getting the data for calculations easy to compute?

That is actually a question with many different answers. The reason that I say that is there are many ways to calculate VaR. If model your positions using the normal distribution then it's not a good measure in any cirtumstance. If you introduce a GRACH process and t-distribution then you are starting to get closer, you will have a good measure for the long positions. But, I think that you have to use some kind of hybrid distribution (2 distributions sandwiched together, in that case you can model stochastic volatility, skewness and kurtosis) if you want to get a reall "good" measure of risk from a VaR. Now, if you really want to get serious, look into expected shortfall.

Why not implement a historical simulation methodolgy coupled with some rigorous scenario analysis?

Also see Mark Kritzman's paper: http://www.windhamcapital.com/workingpa ... _Short.pdf

I think ( if I don't get it wrong) that your question reduces to the question of whether VaR is a good risk measure in general. Then you can get your answer from the literature. It appears that assuming a normal distribution could lead to very inaccurate results and that other fatter tailed distributions could give better estimates of risk. However, if you use weekly or monthly returns the empirical distribution is closer to the normal than when you use daily returns. Thus, in this case the normal distribution could give you satisfactory results. Regarding the historical simulation there is an issue with the number of returns available for VaR analysis. The empirical distribution is very discrete at the tails. Thus, it is very easy to significantly underestimate or overestimate risk. Also, you should considere the different points of view of regulators and investment banks. What could be good risk estimates for regulators might not be good for IB and vice versa. Maybe a good idea is to use Lopez type backtesting techniques to assess your VaR estimates. In summary, I don't think you have a straight forward and simple answer to your question. It depends on so many things like confidence level, frequency of returns, time period, perspective, sectors, particular stocks, aims (reporting purposes or setting trading limits).....Hope that helpsKostas

Besides all the standard problems with VaR already pointed out (fat tails etc), the additional problem faced by a long/short equity portfolio in particular (I did not see any particular reference to a specifc asset class) is that one almost always relies on a factor model of some kind (Barra/APT among others). The adequacy of VaR then depends crucially on the ability of the factor model to pick up the basis risk in the portfolio. As an example, a single factor (market index, for example) is very likely to lead to an understatement of risk. I thought there was a paper by Tanya Beder and/or Leslie Rahl on the subject, but cannot find the reference. It provides some nice examples of the impact of the assumed factor structure on the measured risk. Hope this helps.

I think I can use var-covar. But then I have to decide whether I want to use equally-weighted historical data, what frequency, how far do I go back. I could use GARCH with skewed distribution. EVT could be useful.I can use historical simulation but there may be too few observations in the wings.If I calculate VaR for individual stocks, I cannot add them up for the portfolio because of the correlation. How would I determine what happens to my VaR if I want to add one more security to my portfolio?If I have 1 US stocks, 1 Japanese stock and 1 Singaporean stock, do I convert everthing in USD? How do I account for FX risk then?