Friday, July 22, 2011

Breaking Down the GLD / SPY Model

I've been researching and analyzing a number of rotation / momentum strategies of late (details potentially to follow), which is one reason why I was so interested in the recent GLD / SPY Rotation Strategy posted by Michael Gayed over at The Big Picture. In a nutshell the strategy attempts to follow rolling monthly momentum to allocate between gold and the S&P 500. He concludes:

Of course, past performance is not indicative of future results, but the simple binary decision of being either long SPY or long GLD depending on which is outperforming the others does seem to suggest alpha can be generated.
While I am not nearly as willing to suggest the framework works (and if it does, it doesn't necessarily work in the manner described), I did think the framework was interesting enough to take a deeper dive.

First of all, let's outline the strategy:
  • Using end of day values for Gold (ETF GLD) and the S&P 500 (ETF SPY), create a price ratio by dividing GLD by SPY
  • If the ratio is greater than the 20 day moving average of the ratio, then allocate to gold; otherwise to the S&P 500
I was able to closely match the results:

What I Like

Before I dive into the issues I have with the analysis, here is what I like...

I like that while the strategy was only allocated to gold about 50% of the days, it still tracked the performance of gold with a correlation of .70 on a monthly basis. That is pretty staggering. Why do I like that? Because it opens up the possibility that it closely tracks gold when gold outperforms and may track equities when equities outperform.


That said, here are my main issues with the model:
  • The data set is very limited
  • The data is from a period in which gold significantly outperformed equities
If you were to have asked me prior to seeing the analysis if I was interested in a model that involved gold and equities that outperformed equities over the past 6 1/2 years, my response would have been a very easy no. Unless the model shorted gold, it would have been just about impossible NOT to outperform equities over that time frame (gold is one of the top performing asset classes since 2004; equities one of the worst).

As a result, I would have been much more interested in hearing about a model that outperformed gold over this time, rather than consistently underperform over its history (20.7% annualized returns vs. 17.9% annualized returns), without much of a reduction in volatility (21.1% vs. 20.3%).

Deeper Dive

Keeping those limitations in mind, lets dive into the idea that gold is a good momentum strategy. The below chart summarizes the performance of the model based on different periods of exhibited "strength" in the model as defined by the level that the GLD / SPY index was above its 20 day moving average and the corresponding one day forward average return in the price of gold and the S&P 500.

An interesting observation is that the only level that gold did not outperform was when the GLD / SPY index was greater than 10% above the 20 day moving average (note that this was less than 2% of all trading days) and GLD / SPY outperformed most when the GLD / SPY index was more than 10% below the 20 day moving average. This:
  1. Indicates the GLD / SPY index outperformed the S&P 500 because gold in most instances outperformed the S&P 500
  2. The model actually shows gold and S&P 500 exhibit mean reversion at extreme levels


I was able to find Gold prices going back to 1992 (the inception of the SPY) over at USA Gold and recreated the model using Gold rather than GLD. The results don't seem too promising prior to the beginning of the gold rally that started in 2001.