Tuesday, November 24, 2009
Given the data in the last post, the temptation to fit some curve is irresistible. The data looked superficially hyperbolic, but that turns out to not be a very good fit. The blue curve above is y = 1/sqrt(-0.00011832x+0.01026645). Of course, I don't have the slightest theoretical justification for this curve, so the extrapolation has no basis beyond guesswork. But, just in case you were curious, it asymptotes at 86.8mbd (but don't forget the data being fit is effectively low pass filtered by the five month moving average, and is an average of three data sources). Probably for entertainment value only...1.
1. It's worth emphasizing for the less mathematical reader than there are always an infinite number of mathematical functions that can be made to fit a set of data like this about equally well, but will do different things outside the range of the data. So unless there is some kind of underlying theory of why a particular model is likely to be correct, this kind of extrapolation should be taken fairly cautiously.