I'd also suggest R. One of the problems with visualizing complex data sets is that, almost by definition, the prepackaged graphics tools don't allow you to create custom-designed graphics that suit the particular data-set you're working with. But with a bit of programming in R you can get amazing results.

There are some R packages that can help too -- I write about one of them, ggplot, here. (Disclaimer: I work for a company that provides support for R.)

I didn't know about that either -- I don't think it's gotten much publicity. A little googling turned up an article on an obscure website. Life is about to get a lot harder for legal green card holders entering the the US.

That's not quite true -- although as a state issue, Prop 8 doesn't have anything to do with this. Like hundreds of other benefits, health insurance has a FEDERAL tax benefit tied to marriage. Even if an employer offers insurance to a same-sex partner, that partner has to pay tax on the full retail value of that insurance, as if it were income. Only a married partner can receive health insurance without the additional tax burden. Because insurance on the retail market is so expensive, the additional tax often makes the insurance unaffordable (as I can attest from experience).

That's one reason the marriage issue is so important to same-sex couples. Many federal benefits are tied up with the act of marriage, in law.

The modified R sources have always been available alongside the binaries at http://www.revolution-computing.com/binaries, and there's now a link on the REvolution R download page as well.

If you want to get a Linux binary, download the sources and compile it. It should work fine on recent versions of RHEL and SLES if you have the necessary toolchain, but YMMV on other variants of Linux. That's the only reason why we currently don't provide free (as in beer) binaries for Linux -- there are so many variants that it's difficult and costly for us to test and support them all.

As an open-source company we support and respect the GPL. We're here to support R and R community, and changes to core R made by our development team (such as 64-bit support on Windows, which we're working on now) are contributed back to the development community via the GPL as they should (and must) be.

That can be a problem (but Google returns meaning results for searches involving R these days).
That said, there's a *lot* of information out there about R. The r-help mailing list in particular is very active (making r-help a good term to add to a Google search).

I maintain a blog about R with the aim of collecting the most useful information in one place. (Disclaimer: I do this as part of my work for a company that provides commercial support for R.)

R isn't just free as in beer, it's also free as in freedom: it's under GPL2. It's the freedom given to all those statistical and programming experts to tinker with R that's made it what it is today. (The use of the word "freeware" by the SAS person seemed like a deliberate slight on FOSS to me.)

There are non-free-as-in-beer (but still free-as-in-freedom) versions of R too: I work for a company that provides support for R (under the name REvolution R) with a model similar to Red Hat Linux. The market share for R is now more than large enough for it to be viable for commercial organizations to support it.

And even that understates the real extreme risk. In your example, $50M is the minimum loss 1% of the time only when the model is correct (more details here). But if the model's wrong (and it is when everything's going to hell), the minimum loss is probably much much larger, as many banks recently discovered.

I would even go so far as to say VaR is a decent number used by the wrong people. From a statistical perspective, VaR is a perfectly decent statistic, given the model's correct. Even if the model's wrong (and all models are), as long as it's measured consistently it's a useful indicator when the underlying financial processes are changing, in much the same way as six-sigma analysis is useful in manufacturing. Part of the problem is that there's often pressure from non-statisticians to change the way it's measured ("stuffing the tails" from the article), or for using it for inappropriate purposes (like in financial statements), or simply persisting on continuing to use it when the model is clearly now wrong (LCTM in 1998, everyone except Goldman Sachs today.)