Saturday, 31 August 2019

HUI to Gold regression revisited

During the gold rally which took us from below $280/Oz by end 2001 to $1000 in March 2008, we were repeatedly told that the HUI miners index fluctuated around half of the price of gold (in USD). It was a trading rule which wasn't really questioned in depth and it seemed to hold for several years.

The mere appearance of that simple linear relationship misses out on the 'leverage' of mining shares to the gold price, which mining investors count on. Yet, using a comparative graph of the HUI index and the gold price, some stretches of miners outperforming gold were identified, especially during the initial phase of the gold bull market. The linear relationship only seemed to set in after the first few years of the gold bull market from about 2004.

As an illustration, you find the synoptic long term HUI and Gold graphs on the below figure, with gold (red line) on the left axis running to 2000 and the HUI on the right axis running from 0 to 1000. The axis choice reflects the linear relationship designed above. It is evident that it only holds over a short stretch of time from about 2004 onward till early 2008. The HUI index started 2004 at 245 with gold at $415. For your information: the HUI quotes at 228 on Aug 30, 2019, with gold now over $1500. Always useful to look at the past in order not to feel overexcited about the present !

Price of Gold in USD, left axis in red, HUI right axis in blue

Yet, without any idea on what was to follow, enthusiasm didn't seem out of place as gold first hit $1000/Oz in March 2008. From summer 2003 till March 2008, the HUI = 0.5 * Gold linear approximation seemed to uphold pretty well.
Price of Gold in USD, left axis in red, HUI right axis in blue
After identifying the HUI / Gold regression relationship that remained valid from 2012 till 2017, it was considered useful to repeat the exercise on earlier data.

Regression between HUI and Gold price during the 2001-2008 gold bull market 

Seeing the gold price more or less flat during the initial recovery of miners from their abysmal gold bear market low, we attempt checking for a regression from mid March 2002 (with gold at $290/oz) till 14 Mar 2008 when gold hit $1000/Oz for the first time. The slope and intercept found were such to relate HUI against Gold as:

HUI = 0.6183 . (Gold - 97.7) with a correlation of 0.946 over the target time interval (6 years)

Especially since the HUI remains far below half the gold price throughout 2002 and well into 2003, the above regression relation fits far better than does the empiric simple linear trading rule. Moreover, it brings back 'leverage' into the equation, with HUI percentage gains more pronounced, especially during the first two years when the intercept was a substantial fraction of the gold price.

Near both ends of the time stretch, true HUI observations fall below the regression values, which more or less indicates that we could have done better with further restriction (somewhat at the cost of credibility however).

Eliminating time, you get the regression line between HUI and Gold (observations are red dots).

Regression line and observations on HUI and Gold from 15 March 2002 till 14 March 2008
(the first time gold closed above $1000/Oz on Comex)
When implementing the above regression to a synoptic graph of the HUI and the gold price, we obtain following:
Synoptic view of the gold price (left axis in red) and the HUI (right axis in blue)

This synoptic graph shows the gold price on the left hand scale (red line), which also is the regression value of the HUI when read at the right hand scale. The real value of the HUI is shown in blue on the right hand scale.

The HUI outperforms relative to the gold price at the start of the observation interval as the blue HUI graph rapidly regains the red (gold price). However only after mid 2003 the HUI really leaves the gold price behind. Outperforming gold ends after the first quarter of 2004, with a minor HUI excess rally during autumn that year. From 2005 till autumn 2007, tracking errors go both ways and remain balanced. It should be borne in mind that the 'tracking error' means more than just outperforming (when rising) or lagging (when sliding). The regression already captures the average non-proportional response of the HUI relative to the gold price.

After gold broke its 1980 all time high above $800, the HUI corrected and the tracking error turns negative again. Little did we know that this degradation was the first crack in a system that was about to fall apart.

* * *

With the benefit of hindsight: I retrieved one article of 2007 which is following a similar approach:
'Understanding the Relationship between the Gold Price and HUI' (Seeking Alpha, Thomas Tan)

And (as always) the most important danger of a model is sticking with it after its expiry.

* * *

Further reading on this topic

No comments:

Post a comment