My wife and I are in the process of hiring a financial planner. While I think this is important to do, our search has brought me back into vivid contact with some of what I consider the nonsensical jargon of academic finance. I want to write about the general idea of “non-correlated assets,” but I’m going to start by writing about beta.
In the early days of computer-driven finance, just after WWII, economist Harry Moskowitz proposed beginning to assess the risk of a portfolio by analyzing the interrelationships among individual stocks in it. That task proved too daunting for the computers of the day for anything but small numbers of stocks. Others suggested correlating everything to one standard, an index like the S&P 500, for instance, instead.
The regression that would do this has the form of y = α + β(S&P). This is how beta, the correlation between a given stock’s price movement and that of the market, was born.
So far, so good.
…and gold stocks
One day, people discovered that there was a class of stocks–gold stocks, in particular– that had a beta of 0. This spawned the idea, encouraged by the gold-bug prejudices of the day, that one could lower the beta of a portfolio just by adding gold stocks. One could add, say, technology stocks with a beta =2 and offset the risk by adding gold stocks in the same amount. Simple math said the combination had a beta = 1, or risk exactly equal to that of the market.
Some institutional investors actually bought the theoretical argument about the “magic” property of gold and altered their portfolios in the way I just described.
By doing so, they exposed themselves to the 20-year bear market in the yellow metal that lasted from 1950 to 1970. They lost their shirts.
They realized only afterward that a beta of zero did not mean that the asset in question had no risk. It meant instead only that the zero-beta asset did not rise and fall in price in line with the stock market. In this case, the “uncorrelated” price went straight down during a period when the S&P gained 500%+. So much for non-correlation.