Flirting With Models

These might be of interest to some folks reading this blog…

Another great and free Haskell resource can be found here.

A great little Haskell tutorial for those of you who liked why’s (poignant) guide to Ruby is Learn You A Haskell. Perhaps not as zany, but also easily digestible (especially if you have previous functional programming experience).

When playing around with Artificial Immune Systems for classifiers, I stumbled across a more interesting measure of distance than the typical Euclidean measure called the Mahalanobis Distance, which incorporates the correlation of data set.

Optimized Firefox is fast like woah. Still got a PPC? No worries. Intel/PPC builds available nightly here.

I wonder if Fairfield Greenwich ever noticed that their image had them shifting their normal curve to negative skew? Nothing like saying you will keep the left tail and cut the right…

Solver was removed from the latest version of Excel on Macs. FrontLine systems provides the solution. It is an external application that links to the current open sheet. Works very well, so check it out!

If you haven’t caught it yet, Thrust version 1.1 is out. Check it out here.

Doing some Macro homework. I keep stumbling across the same problem over and over: economists who love to forecast way into the future — and way beyond the range of their sample.

For example, reading “America’s Dark Materials” in the Economist, we find the sentence: “After making $36.2 billion in 2004, America made just $4 billion on its net foreign assets in the first three quarters of 2005. If it continues on its present trajectory, it will shell out about $190 billion in 2010, Mr. Cline calculates.”

So let me get this straight — two data-points over three quarters, Mr. Cline was able to forecast out 6 years? Yes, that seems totally reasonable.

The same goes for BRIC growth rates. Sorry — I simply don’t buy that growth rates over the last 5 years can predict the next 40.

Rule of thumb: the length into the future of your prediction should be smaller than your sample size. Preferably, much smaller.

Max Dama puts up some interesting notes on the “Curse of Dimensionality”.