Pairs trading gold

Over at the Mechanical Investing board, a user mentions pairs trading gold. I'm working on some new tools, and this gave me an excuse to run a simple test. Since GLD (a gold ETF) and/or GDX (gold miners ETF) has a fairly short history, I decided to test extending the history somewhat, creating a proxy for GDX and testing the ratio with the actual gold price. The goal with the proxy and gold price is to allow for a longer backtest.

For the GDX proxy I picked the US listed companies currently owned by the GDX ETF: AEM, ABX, AU, GFI, NEM, HMY. Comparing this proxy with GDX gives the following results:

GDX and proxy

At least over the period that the GDX ETF exists, this proxy isn't too shabby. But just to be on the safe side, let's compare the shape of the two ratios, GDX/GLD with proxy/goldprice:

GDXGLD ratio
Gold proxy ratio

Based on this, I would say the proxy works quite well. Since the proxy has a longer backtest, how does that look?

Gold proxy ratio long backtest

By reducing the GDX proxy to the companies AEM, ABX, GFI and NEM, it's possible to backtest back to the beginning of 1990, as shown below:

Proxy ratio long

Note to myself - Code used for generating graphs:

# plotparams(size=640x480)
# cquotes(19900101,20080410,ya:AEM,ya:AU,ya:ABX,ya:GFI,ya:HMY,ya:NEM,pi:IDXDATA/$GOLD,ya:GDX,ya:GLD)
# createscale(GDXproxy,ya:AEM,ya:ABX,ya:AU,ya:ABX,ya:GFI,ya:NEM,ya:HMY)
my $sum=0;
for (@d[1..$#d]) { $sum += $_; }
$sum /= @d;
return $sum;
# plot(GDXproxy,ya:GDX)
# create(proxyratio,GDXproxy,pi:IDXDATA/$GOLD)
return $d[1]/$d[2];
# rawplot(GDXproxy)
# plotparams(fromdate=20060523)
# rawplot(proxyratio)
# create(GDXGLDratio,ya:GDX,ya:GLD)
return $d[1]/$d[2];
# rawplot(GDXGLDratio)
# plotparams(fromdate=19700101)
# rawplot(proxyratio)
# createscale(GDXproxyLong,ya:AEM,ya:ABX,ya:GFI,ya:NEM)
my $sum=0;
for (@d[1..$#d]) { $sum += $_; }
$sum /= @d;
return $sum;
# create(proxyratiolong,GDXproxyLong,pi:IDXDATA/$GOLD)
return $d[1]/$d[2];
# rawplot(proxyratiolong)