I’ve got a portfolio of stocks and I’m trying to model the 95% and 99% VAR, and I have no clue where to start. Anyone with any experience?

You need to make some assumptions for each stock. People often just use historical data: 1) Expected return 2) Standard deviation of returns 3) Covariance of returns between each stock Then, just plug everything into the covariance formula and find the return that is lower than the 95% or 99% worst cases. That’s about it.

Even easier than that is just pretend that you held the portfolio for the last 1000 trading days and see how it would have performed. The 10th worst is the 99% VaR and the 50th worst is the 95% VaR. ohai’s is the parametric VaR and this is the historic VaR (I would actually use ohai’s, but then I couldn’t write this…).

Joey I think I’m just gonna do it your quick and dirty way. Any way easier than pulling up 1000 data points for each stock on Yahoo Finance, and then calculating portfolio returns, to get those returns?

For free, Yahoo Finance might be your best bet. Unless you want to learn some quick and dirty R programming using the quantmod library…

I wouldn’t assume constant volatility. Even Risk Metrics approach is better than the ones described so far. I also wouldn’t worry about mean return because it’s probably very small compared to volatility. http://en.wikipedia.org/wiki/VaR http://en.wikipedia.org/wiki/RiskMetrics

^ I respectfully disagree. The guy has no idea how to start and you’re sending him off an some exponentially weighted covariance calculation? (btw Allan Malz was the RiskMetrics guy in charge of coming up with all that stuff and if you ask him you wouldn’t get anything more enthusiastic than “well it’s at least as good as all the more complicated methods we tried over the range of all the traded factors in the world”). Since we can’t even agree about whether variance is finite, does it really matter how well we account for non-stationarity?