In the Quiet of August, a Troubling SpikeThursday September 08, 2016 12:48
Are gold, copper and oil signaling a third wave of commodity deflation?
Gold Country, Diamond Mountains - Eureka, Nevada (photo courtesy Mariana Titus)
September 8, 2016
A docile August
The S&P 500 made an all-time high in August but remarkably scored less volatility than key commodities and major currencies – about 1/3 the volatility of gold price. Gold attempted $1,373 per ounce early in the month then swung perilously close to the key $1,300-level as August closed. Gold is now grazing in mid-$1,350 pasture. Copper continued its July downtrend of lower highs nearly breaching the important $2 per pound-level at month close - about 1.9x more volatile than gold. Presently, the red metal attempts a meek recovery at $2.1 per pound. Unsurprisingly, crude oil led the group with 4.4x the price wiggle of the yellow metal bouncing $9 per barrel from early month lows to nearly $50 per barrel. Presently the gooey stuff is finding some comfort around $45.
All-in-all a pretty docile August – or was it?
A troubling spike
Figure 1 illustrates a divergence indicator derived from 1- and 3-month rolling correlations of Comex copper and Nymex crude oil (WTI) with respect to gold. Over the last 10 years, spikes close to 2.0 occur quite infrequently but before periods of memorable volatility and market turmoil.
Figure 1. A divergence indicator flags a warning
On August 29, this divergence peaked to a level near the onset of the 2008-2009 crisis and also during Arab Spring 2011. In the current case, gold is negatively correlated with WTI on both a short and longer term basis and positively correlated with copper by the same comparison.
Correlation (Rho-) maps, a powerful tool
Correlation (Rho-) maps are a powerful tool to understand these commodity correlations with gold and key divergences when they occur. A rho-map is simply a scatter plot of rolling correlations with the Y-axis representing a 3-month period; and the X-axis, a 1-month basis. Figure 2 shows a 10-year record of gold and copper, and gold and WTI data within +/- 1.0 correlation boundaries.
Figure 2. Rho-maps for copper and WTI with respect to gold (October 2006 to the present)
Some characteristic patterns emerge like constellations in the sky. Notice how the stars, or correlation points, bunch up in the positive upper-right (+/+) quadrant and are less dense in other quadrants. This indicates that short (1-month) and longer term (3-month) correlations tend to be positive for both copper and oil. The red metal is in the +/+ quadrant with gold 56% of the time and for oil, about 51%. Negative correlation in the opposing -/- quadrant is much less frequent: 16% for copper and 12% for oil. There are fewer points in the transition or mixed correlation (-/+ and +/-) quadrants too.
Gold price movements are then in step with copper and oil more often than not. The stellar center-of-mass, or average correlation over 10-years, is in a positive range of 0.2 to 0.3 (denoted by the red triangles in Fig. 2). Markets become more interesting when key commodities stray from such “normal” behavior.
Three cases of extreme divergence, a fourth not
Tom McClellan, author and editor of the respected McClellan Financial Publications, reviewed Figure 1 and asked a very good question. If the extreme spikes herald coming market chaos, why didn’t a spike occur before the August mini-crash of last year?
To answer his question, I reviewed the four cases summarized in Figure 2. The rho-map of this figure shows the correlation data indicated for each case. Divergence is the distance between data points of different correlations – for this analysis, correlations of (gold, copper) versus (gold, WTI).
Figure 3. Rho-map for four case studies
My conclusion is that the mini-crash (case #4) was transitory whereas the extreme divergent peaks of this August (case #1), pre-Great Recession (case#2) and Arab Spring (case #3) are precursors to longer term turmoil in the commodity space.
The mini-crash was triggered by a surprise Chinese yuan devaluation and caused market panic: gold popped while equities and commodities plunged. For case#4, WTI and Cu are both negatively correlated with gold in the short term and thereby demonstrated little divergence (div=0.163). This was a "Chicken Little" event in that once market participants realized the sky wasn't falling, prices recovered and the gold rally faded.
For the other three cases, gold is positively correlated with one commodity and negatively with the other. For Case #1, gold is positively correlated with copper and negatively with oil. Case #2 and #3 are flip-flops – gold is positively correlated with oil, negatively with copper. It is noteworthy that the corresponding data points are near and on opposite sides of the unit circle. This explains the importance of peaks near 2.0 - the diameter of the circle (note 1).
Data points for Case#4 are also near the unit circle but in the same quadrant. Divergence is thereby small with short and longer term correlations of opposite signs suggesting the transitory nature of the event (note 2).
With respect to copper prices, Goldman Sachs forecasts the possibility of $4,000/tonne ($1.81 per pound). Sub-$1.50 copper occurred in 2008 until reflated by QE1. Sub-$2.00 copper was briefly visited after Arab Spring and the U.S. debt crisis in 2011. If Goldman Sachs is correct, all three copper lows will be preceded by divergent peaks ~ 2.0.
Oil followed a similar fate as copper in the past and there are plenty of arguments for another downturn in the coming months. The broader Bloomberg Commodity Index (BCOMTR:IND) has been trending down since early July. Comparing the present with 2011 and also the 2008-2009 crisis may signal that we are entering a third wave of commodity deflation.
Let’s hope not.
From the heart of North American Gold country – Cheers!
The Eureka Miner (http://eurekaminer.blogspot.com/)
Note 1: the theoretical maximum divergence is the diagonal of the correlation +/- 1.0 box or 2.828 suggesting 2.0 is indeed an extreme level.
Note 2: the theoretical maximum divergence for data points in the same quadrant is 1.414, thereby less than 2.0