Gold analysis part 2, a question for the social scientists …
For an industry awash in data, the empirical financial modeler faces an interesting problem — the quality of the data is always suspect. For instance, say you want to rank, by country, the percentage (of weighted cost) of non-performing real estate debt. Based on the published statistics, the ranking would be China (lowest), followed by Germany, followed by the U.S. The pure mathematical modeler among us will be disappointed to learn that the “true” ranking is probably Germany, U.S., and China.
A healthy skepticism about modern financial statistics is a modeler’s best friend. The qualitative descriptions buried in infotainment (German real estate culture described by this Vanity Fair piece by Michael Lewis on the German economy) and the academic literature (see this interview with a Northwestern professor on real estate loan fraud in China) provide us with instinct about how people lie with statistics. Without arguing the finer points of accounting standards, even the one ‘true’ statistic — market price — has come under attack in flash market crashes lately.
Given a global economic environment where statistical financial signals can deceive, it is with great interest that I read about Statistical Arbitrage (stat arb) strategies for betting on gold prices. The logic goes that if we know the underlying factors that contribute to the price of gold then we can bet (in sub-minute to sub-day intervals) on changes to the price of gold based on these underlying factors because the price of gold will tend towards the mean of the weighted value of its factors.
The paper that I linked to in Part 1 of this discussion contains a possible model for the factors for gold and the statement that our mission is to evaluate how to trade for gold, given the available information.
The model presented in the paper can be summarized and empirically tested for just this year using minute-by-minute pricing data from ETFs (CPI, GLD, VXX, SPY, and VGK). On first blush, the results from the statistical analysis appear promising:
If the true factors for Gold have been identified, the Statistical Arbitrage of gold is quite possible. We simply wait for differences in pricing among gold and/or its factors and then buy or sell the under or over-priced security. Now the question for the social scientists. What are the problems with this approach?
