Gundlach indicator: copper, gold & treasury yields at a tipping point
Kitco Commentaries | Opinions, Ideas and Markets Talk
Featuring views and opinions written by market professionals, not staff journalists.
Balance sculpture near DoubleLine Capital LP Headquarters
California Plaza, Los Angeles, California
A “fantastic” indicator of interest rates persists
October 14, 2019
I have written a number of Kitco commentaries on the stunning relationship of the U.S. 10-year Treasury yield with copper and gold prices. The following is an update to my previous column,
Gundlach Indicator: Treasury Yields and Copper-Gold Ratio Plummet (September 3, 2019). It is worth repeating the origin of this relationship and behavioral character:
Jeffrey Gundlach, CEO of DoubleLine Capital LP (DoubleLine®), noted in his 2017 forecast that the copper-to-gold ratio was a "fantastic" indicator of interest rates. High-fidelity yield models based on this relation are often possible on a short- and even longer-term basis. During times when the copper-gold ratio does diverge from the 10-year yield, the rise or fall of the former usually portends a rise or fall of the latter.
Jeffrey Mayberry, Co-Portfolio Manager of the DoubleLine® Strategic Commodity Fund, has studied the relationship in great depth and makes this important observation:
The ratio’s absolute level is irrelevant. What matters is its direction – and whether the yield on the 10-year Treasury moved in the same direction or diverged. In past episodes of divergence, the 10-year yield has eventually tended to follow suit of copper-gold (“The Power of Copper-Gold: A Leading Indicator for the 10-year Treasury Yield,” Jeffrey Mayberry, DoubleLine® Funds)
Starting this spring, Treasury yields and the copper-to-gold ratio plummeted and bottomed in early-September. Since then, both have advanced higher, retreated lower and reversed higher again last week.
How has the Gundlach Indicator performed over this volatile period?
Since mid-June, weekly regression models of 10-year Treasury yields based on the copper-gold ratio have proven very accurate. The estimation errors are in a range of 5-7 basis points (bps) with a “goodness-of-fit,” or R-squared, near or better than 0.9. These models also provide valuable statistical upper and lower bounds for yields.
What does the indicator tell us going forward? Looking at four charts provides some clues. First, the latest model:
High-Fidelity 3-Month Yield Model
Figure 1 is a 3-month regression model of 10-year U.S. Treasury yields based on the ratio of copper and gold prices from July 16, 2019 through Friday’s close, October 11, 2019.
Figure 1 – U.S. 10-year Treasury yield model based on Comex copper & gold prices
Comparable to the weekly models that preceded it since June, the statistical error is just under 6.5 bps with a respectable R-squared of 0.8934 (I consider 0.85 to be a threshold for developing useful models).
Importantly, the model provides upper and lower 2-standard deviation bounds for yields. The upper bound suggests yields approaching 1.9% are possible in the near-term with a floor above 1.6%. Friday’s yield estimate is within 1 bps of the actual yield (1.7618% vs. 1.7520%).
Observe how quickly yields rose after the September 4, 2019 low of 1.454%, then reversed lower, and are now on the rise again.
Correlation Map Indicates Strong Persistence
A correlation map (rho-Map©) is a powerful technique for detecting changing directional behavior between two market variables. Figure 2 is the 3-month (Y-axis) and 1-month (X-axis) rolling Pearson correlations of the copper-to-gold ratio (CGR) and the U.S. 10-year Treasury yield.
Figure 2 – Copper-to-Gold Ratio & 10-year U.S. Treasury Yield Correlation
The correlation trajectory starts at the beginning of the current 3-month model data, July 16. Each point represents a market-day; the first starts in the upper-right quadrant of positive correlation (shaded area). The trajectory, or sequential connection of data points, quickly enters the upper-left quadrant as the ratio begins to de-correlate from yields on a 1-month basis.
However, this direction dramatically reverses Friday July 26, 2019 closing a week of market turbulence - the election of U.K.’s Boris Johnson, heightened fears of a “hard Brexit” and tumbling pound sterling. Additionally, the European Central Bank hints key interest rate cuts for the first time since early 2016 in response to a concerning economic slowdown in the region. On August 23, President Trump threatens in a tweet to escalate tariffs on Chinese imports and demands companies to cut ties with China. The dog days of summer become anything but tranquil.
In such periods of market turmoil, a “risk-off” sentiment typically drives demand for U.S. Treasurys and gold. Bond and gold prices rise as copper prices fall. Accordingly, the CGR declines with yields (which are inversely proportional to price) reinforcing positive correlation between the two.
Starting July 26, the trajectory quickly accelerates to high positive correlation (arrow “A”) as evidenced by the increased spacing between data points. When short- and longer-term correlations have the same sign, we say the relation exhibits “persistent” correlation. The shaded upper-right quadrant indicates positive persistence. Data concentrated in the +0.8 by +0.8 box indicates “high persistence density” (darker shaded area) fortified by the August 23 tweet.
As evidenced in Figure 1, construction of a high-fidelity yield model is possible for cases of positive correlation persistence with improving accuracy given high persistence density.
In terms of Mr. Mayberry’s definition, the relationship moved from a brief flirtation with divergence (July 16-26) to a strong directional behavior between interest rates and the copper-gold ratio – a correlation journey that continues to the present.
Figure 3 illustrates this journey in terms of increasing positive persistence (P+, blue trace), persistence density (magenta trace) and improving model R-squared (diamonds spaced at 1-week model updates).
Figure 3 – Correlation Persistence & Model Goodness-of-Fit (R-squared)
Positive persistence, expressed as a percent, is simply the number of points with positive 1- and 3-month correlations divided by the total data points for the period of interest. Similarly, persistence density is equal to the points in the +0.8 by +0.8 box divided by the total points. For this analysis, the time interval is 3 months or 63 market-days.
We note that R-squared rises from a near-zero value May 10 to greater than 0.85 by June 14 – a rapid transition from a useless to a quite accurate regression model. Concurrently, positive correlation persistence and density increase taking a leg up after the July 26 reversal of Figure 2 (arrow “A”). As of Friday, persistence is a strong 84.1% with a high density of 55.6% (I consider densities above 50% to be high). Although R-squared has declined some, it is still at a respectable 0.89337.
Will the Gundlach Indicator maintain such fidelity or is divergence on the horizon?
Copper & gold correlation
Figures 1-3 offer no strong evidence that the present relationship is in trouble. However, there are some hints of impending weakness: by Friday’s close, the 1-month correlation of Figure 2 is outside the high-density box, goodness-of-fit has fallen some from earlier values this summer (Figure 3) and estimation errors are approaching the top of the range from mid-June (6.5 versus 7 bps).
We need to look a little deeper to determine if this is just statistical noise or the beginning of divergence. The correlation map for copper and gold shown in Figure 4 offers insight.
Figure 4 – Copper-Gold Correlation
As in Figure 2, the copper-gold correlation trajectory begins July 16. It is in a transitional quadrant of positive 1-month and negative 3-month correlation. After the July 26 reversal, the trajectory rapidly enters the lower-left quadrant of negative correlation (i.e. arrow “A” enters the shaded area). Over three months, this negative persistence (P-) is quite strong at 66.7%. In other words, gold and copper prices are mostly going in different directions supporting the “risk-off” sentiment described above (gold up, copper down). Notice this has changed recently with the trajectory leaving the persistent quadrant after October 1. Although Friday’s data moved closer to the lower-left quadrant, it remains outside (dotted ellipse).
History provides a guide to what may happen next. In calmer markets, gold and copper tend to be positively correlated. Over the last seven years, they demonstrate positive persistence 50% of the time and negative persistence only 17%. The remainder of the time the two metals wander in transitional quadrants (33%). By this measure, the 66.7% negative persistence of Figure 4 is elevated historically if not aberrant.
Furthermore, the copper-gold ratio appears ready for further expansion. On September 3, the CGR bottomed at 0.162 or, in reciprocal, an historically elevated 615 pounds per ounce. Friday closed at 0.177 or 566 pounds per ounce indicating that expansion is underway.
Putting the pieces together
Copper, gold and Treasury yields appear to be at a tipping point. The CGR direction and copper-gold correlation map should be monitored closely to divine future direction for all three.
Clearly the negative persistence of copper and gold correlations contributed to the high positive persistence and density of the Gundlach Indicator over the last several months. This could fade quickly if last week’s mini-trade deal with China leads to improved trade relations and/or a hard Brexit is avoided. Treasury yields will likely still follow the CGR but with less correlation intensity. As a consequence, yield model accuracy will suffer.
Another possibility is divergence of the two with yields going their separate ways. As Mr. Mayberry has pointed out, CGR direction will still be important to monitor as a leading indicator for yield direction. However, sustained divergence precludes any meaningful yield modeling based on copper and gold prices.
If a greater U.S./China trade deal and graceful Brexit prove chimeras, there could be a repeat of this summer’s strong negative correlations of gold and copper. As mentioned above, this outcome does face some historical headwinds.
My “risk-on” bet is slowly rising yields and CGR as the latter reverts in the direction of historical norms (greater than 0.185 or less than 540 pounds per ounce). The near-term yields appear capped at 1.9% but may creep above 2% before 2020. Copper should find comfort above $6,000 per tonne ($2.72 per pound) in a “risk-on” scenario.
However, there are enough goblins in the dark corners of domestic politics, world trade and geopolitical tensions to secure a solid price floor for gold around $1,450 per ounce. A reignited gold rally and falling yields before Halloween is certainly a possibility if any of the recent witches reappear at the global door step. If there is a rush for safe havens, Comex gold will likely fail to reach $1,600 per ounce in 2019 and 10-year yields will remain above the September low.
Hats off again to the DoubleLine® team for discovering this important market indicator with proven performance in troubled markets. The market events since spring provide continued testament to that record.