Saturday, October 3, 2015

Should we consider Gold?

I read a post on Mebane Faber’s blog a few months back about something he calls the “Three-Way Model.” It’s a simple trend following model that uses 3 ETFs: Stocks, Bonds and Gold. The rule is simply: Invest equally in whatever is going "up" (ie. 3 month moving average > 10 month moving average). The results over the past 40 years have given a 12% annualized return (see below).

This is not nearly as good as GEM’s 18% compounded annual return over the same 40+ year back-test period. Nevertheless, this got me wondering: What would happen if we added gold to GEM’s relative momentum engine?

In a previous post, Markets are Interconnected, we saw that GEM already indirectly profits from US Dollar and commodity cycles when it switches back and forth between US and non-US stocks. But curiosity struck me and so I took a look at adding gold to the model. The below chart shows this modified model (in blue) compared against the original GEM (in red):

By adding gold to the model we get a compounded annual return of 21%, up from the original model's 18%. Over the past 43 years, the modified model would have taken a mere $100 and turned it into a staggering $400,000!

Now before we go running out on the streets of Syracuse naked shouting “Eureka!” let’s take a closer look. 

First, the modified model's higher returns came at the cost of higher volatility. The standard deviation of the modified model shot up to 18%, which is up from 13%. So the Sharpe Ratio didn't improve that much at all. Second, the modified model made 40% more trades over the 43-year backtest period.

Third, and most important, is that all of the modified model's higher returns were concentrated in the '70s. In 1971, President Nixon took the United States off the gold standard. This decoupling from the US dollar led gold to quickly rise to a level determined by the market, instead of being artificially suppressed.

Take a look at the chart below. It shows the modified model (blue) and gold (purple). From 1971-1980, gold increased almost 18-fold! If we remove this period from our backtest, we actually see that the modified model produces lower returns yet has higher volatility than the original GEM. 

Finally, below is a distribution of monthly returns for both the original and modified model as well as the S&P 500.

We can see that the original model (red curve) produces its mean return with greater consistency than both the modified model and the S&P 500. Furthermore, the original model doesn't rely on just a few strong outlier months like the modified model.

Hence, we can conclude that the original GEM is the best. Without the one-time boost that gold received in the '70s, the original GEM offers higher returns with lower risk. Furthermore, the original GEM already profits indirectly from cycles in the US dollar (and hence commodities like gold) due to its switching between US and non-US stocks. 
I think this post is a great example of why you should not blindly strive to optimize backtest results. You want a model that will do well even in the absence of special one-time boosts.  

Results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Additional information regarding the construction of these results is available upon request. Past performance is no assurance of future success. Please see our disclaimer page for more information.