My STFU thesis

I'm about done with my dissertation here at ScrewTape Files University (STFU). I've actually been sitting on it for awhile, patiently awaiting Lewis Cypher's return from one of his bi-monthly pleasure jaunts to Thailand so he can look at the damn thing and pass it on to the graduating committee. (He owns a villa in Nonthaburi out of which he runs a business of some sort. I think it's a gelato joint. I tried google earthing the address he gave me, but all I see are police cars, so that's got to be wrong.)  Anyway, I thought I'd post my work in progress here for Screwtape readers to comment on. It's called "The Second Principal Component of Multiple Fibonacci Moving Averages, Medians, Maxima, and Minima as leading Price Indicator."

In short, every day gold has dozens of numbers associated with it: the X-day moving average, moving maximum, moving minimum, etc., where X can be any Fibonacci number (I've limited myself to 8,13, 21, 34, 55, 89, 144, 233).  For example, check out the 233-day moving minimum, which captures gold's "floor" pretty well. (Data from GLD, hence starts middle of 2005)

 So, all those Fibonacci indicators convey a lot of information, but how much of that info is meaningful vs. noise, and how much of it is redundant vs. unique? 

The first principal component (PC) of all those indicators collapses dozens of numbers into one number (per day), namely a number that extracts the most information from them by means of the magic of mathematics. The second PC does the same thing as the first, only it's value is extracted from the first PC's leftovers (hence they're independent). 

We see below that the first PC tracks the thrust of price over time (linearly transformed here to be on the same scale as GLD). In a sense, the black line is the "fundamental" moving average, distilling all that's important from many other related and frequently used indicators. Indeed, it would appear that buying above the black line is not wise. You can also think of the distance of daily price from the black line as the "Turd Troll-to-Thanks ratio," as gold bug sites probably get lots of thank you's in proportion to how far above, and trolls in proportion to how far below price goes.

More importantly, the second PC appears to function as a momentum indicator of some sort. (It is clearly correlated to the residuals between GLD and the first principal components linearly transformed to fit it). I didn't clutter the graph with vertical lines, but it's easy to see that the tops of the indicator correspond perfectly to bottoms in price, and that when the indicator clearly rolls over, it does not change direction without a sharp increase in price first.

If gold drops precipitously from here without a strong move up first, that would be unprecedented. Though, like the proverbial tree falling in the forest, only because the pattern has now officially been discovered.