Outlook For Gold And Gold Equities In 2016

|
Includes: GDX, GLD
by: Dennis Boyko

Summary

Gold prices exhibit long-term trends which are masked by short-term variance in price (i.e. gold price has a fractal nature).

Linear regression of gold prices over multiple time periods can be used to smooth short-term variance and predict the long-term trend with much better than even odds.

Gold trend as of December 2015 remains firmly down and odds are high that the down trend will continue in 2016.

In the next several weeks to a few months, the odds favor gold price rallying above the multi-quarter regression line creating an excellent short setup.

Stated in terms of odds, the end of year 2015 outlook for 2016 is as follows:

  • Odds are 16 in 20 that 2016 will see a continuation of the current gold down trend.

  • Odds are 2 in 20 that 2016 will end as a well-defined up trend year.

  • Odds are 2 in 20 that 2016 will not have a well-defined trend.

The body of this article outlines the multi-quarter regression model that forms the basis for prediction and goes on to show the derivation of the odds above.

I plan to post a follow-up article when the timing is first right for a short trade in 2016. The main ideas are as follows:

  • The setup for the ideal short trade in the gold sector in 2016 will occur when the gold price has moved well above its the 6Q gold regression line (as computed at that point in time).

  • When the "well above the 6Q regression line" is hit (a fuzzy criterion), the odds for a profitable short trade (short GLD, GDX or your favorite "weak" gold equity in GDX or any major gold ETF) will have odds approaching:

    • 18 in 20 chance of being a winner.

      • Trade should be good if the down trend continues in 2016 (odds 16 in 20).

      • Trade should also be good if 2016 ends without a well-defined trend (odds 2 in 20) since a short when prices are well above the 6Q regression line will still likely be a winner since gold prices trade in a range and your entry would be at or near the top of the range.

      • Trade will like go badly if 2016 turns ends in an up trend (odds 2 in 20) - see 1992 forecast for 1993 and 2007 forecast for 2008 in the per-year trend charts at the end of the article).

      • Odds of the trade winning will be equal to 1 minus the odds of the trade going badly, or 18 in 20.

    • Caution: picking a point near the top of a gold price move above its 6Q regression line is a mix of art and statistics (remember that the regression line has a look-ahead bias). For some of the years in the 1990 to 2014 window, picking a top above the 6Q regression line would have been tricky. As a result the success odds should probably be reduced from 16 in 20. However, even allowing for top picking fuzziness, the odds of success are still well above even money -- I will explore this issue in the follow-up article once the ideal short trade setup is in place.

Alternatively, the keen investor/trader can make odds adjustments easily by reviewing the year by year plots (and reclassifying some of the years from "CORRET" to "WRONG") based on personal trading strategies. I also provide daily updates of the rolling 6Q regression plot at the GoldMinerPulse link provided below.

Finally, odds adjusted for alternate end of regression months periods as well as further back tests from 1970 are presented below (odds are reduced in some scenarios but are always well above even money).

Though not the subject of this article, planning swing trades based on derivations from the multi-quarter regression line may well prove profitable. For example, at the end of 2015, the conditions were ripe for timing long trades because the price of gold was well below the line. Of course when the trend is down, going long when the price falls below the trend will generally not have odds as good as going short when the price rises above the trend -- but even so, the odds can still be better than even money.

Depending on interest in this article, I may also publish mid year or quarterly trend updates.

Motivation and Background: Regression As A Proxy for Trend

gold As the saying goes, "the trend is your friend" and it is clear that corrrectly predicting the trend will increase the odds of profiting from price moves in gold bullion, gold ETFs, and gold equities. Between December 2014 and the end of December 2015, the price of gold fell by approximately 10.3%. The SPDR Gold Shares ((NYSEARCA:GLD)) ETF fell approximately 10.6% in 2015 - by design the GLD ETF tracks the price of gold and does so very well. The Market Vectors Gold Miners ETF ((NYSEARCA:GDX)) fell approximately 26% over the same time window - gold equity ETFs provide leverage to the price of gold moves. Over the last 8+ quarters, the Gold Bugs Index (HUI) had a correlation of 0.96 with the price of gold and trades at a beta relative to gold. The greater leverage can be obtained from long/short investments in individual gold equities. While short-term windows may show some divergence, over the periods of months or longer, when gold falls, GLD, GDX, HUI and gold equities fall. Likewise, when gold rises, GLD, GDX, HUI and gold equities also rise.

The challenge, then, is to identify the current trend and estimate the odds of future changes.

The following observations are based on a review of the plots at GoldMinerPulse Gold Trends Over Years, Quarters and Months:

  • Linear regression against real data tends to produce a line with periods of price above and then periods below the regression line. This is true for real gold for virtually all regression period choices and is a direct result of the way linear regression is computed (i.e. look ahead bias is locked into the regression line computation since future prices are used to compute the line at past points in time).

  • Regression periods of 1 month produce very choppy regression lines, with poor continuity in lines on a month-to-month bias (see charts in the link above).

  • Regression periods of 1 quarter have more consistent quarter-to-quarter continuity but are still relatively choppy.

  • Decade long regression windows, also published at my reference link above, can have large disconnects when major trend changes occur. For example, using a 10 year regression window starting on years ending in 0 (2010, 2000, ...), shows a jump between the 10 year line ending in 2010 and the current decade trend line.

  • Finally the year-by-year regression window in the 1970s to the current period shows very good trend consistency on a year-to-year basis and also pickups the major trend changes efficiently.

Prediction Strategy Summary

It's clear from the outset that the gold trend in 2016 will fall into one of the following categories:

  • Current Down Trend Continues: The multi-year down trend from 2013 continues into 2016 just as it continued from 2013 into 2014 and 2014 into 2015.

  • An Up Trend Starts: It is not impossible for gold to take a sharp up turn in early 2016 and continue rising into 2017. This happened in 1992 at the start of the last major gold up trend.

  • Inflection Year - Gold Price Moves Sideways Without A Clear Trend: By default, if 2016 does not show a clear up trend or a clear down trend then we can say 2016 was an inflection year or a year without a well defined trend.

A review of gold price regressions over multiple time windows in the 1990 to 2015 window can be summarized in the following rule (detailed supporting charts follows):

  • If the multi-quarter trend in gold price remains well-defined at the end of the year, then odds are 16 in 20 that the trend continues for the next year.

  • If the gold price has no clear trend at the end of the year then it is equally likely whether the following year will be up, down or no trend so a prediction is not made.

Gold Price Regression Model For Estimating Trend Odds Regression Window Size

The gold odds calculations are based on the year-end trend computed using gold prices for the previous 6 quarters. The period length was selected based on back test runs over 25 years (and confirmed with a 45 year check) - 6Q had the highest number of correct predictions for the following year and the best results in terms of avoiding making a wrong prediction or more importantly a disasterously wrong forecast (wrong on the trend forecast plus the trend actually reversed in the following year).

#Qs used in
Regression
Total Years #Right
Forecasts
#Wrong
Forecasts
#Wrong AND Trend
Reversed
#FLIP
No Forecast Made
4Q 26 15 2 6 3
5Q 26 17 2 3 4
6Q 26 16 2 2 6
7Q 26 16 2 3 5
8Q 26 15 2 4 5
Click to enlarge

The December 2015 forecast 2016 remains to be determined. Note that
Total Years = 1 + #Right Forecasts + #Wrong Forecasts + #Wrong AND Trend Reversed + #FLIP
#FLIP equals the number of years where a forecast was not made for the following year.

Trend Period Sensitivity

Using the year end as a prediction point is a natural choice and matches market events such as traders squaring positions at year end to lock in profits and preparing for tax filings or long-term investors balancing allocations at year end.

The odds presented above are all based on regression period ending December 31 with a forecast for the following year (January 1 to December 31). It is worth noting that shifting the regression period forward a month at a time results in lower odds with the poorest odds occurring the May/June/July months - perhaps the old wisdom of sell in May and go away is being practiced in the gold market.

My final test on sensitivity involved recomputing the odds year by year starting from December 1970 and repeating year by year through to current time. I also repeated assuming a January, February, March... November regression window close. The odds from these runs are summarized below.

Regression Line
End Month
RIGHT Forecasts
Runs from 1970
RIGHT Forecasts
Runs from 1990
December 25 in 38 16 in 20
January 24 in 41 16 in 23
February 25 in 40 15 in 22
March 24 in 41 14 in 21
April 23 in 39 14 in 21
May 24 in 41 13 in 21
June 23 in 40 12 in 20
July 24 in 40 13 in 21
August 25 in 40 14 in 21
September 26 in 40 14 in 21
October 25 in 38 14 in 19
November 26 in 38 15 in 19
Click to enlarge

The 6 quarter regression window and correlation threshold of 0.22 were tested in the above runs but significantly better choices were not found in 100+ alternate combination spot checks.

Decision Rules

If the regression has a correlation in the range of -0.22 to 0.22, then no prediction is made for the following year. In the year-by-year charts below the prediction for the following year is shown as FLIP and the prediction result is NA.

If the 6 quarter regression correlation is >= 0.22 then the prediction for the following year is UPTREND. If the slope of the gold price regression for the next full year (year for which the prediction was made) is positive with a correlation >= 0.22 then the prediction is marked as CORRECT and WRONG otherwise, with a possible refinement of WRONG TREND CHANGED if the absolute value of the full year correlation was better than 0.22.

If the 6 quarter correlation is <= -0.22 then the prediction for the following year is DOWNTREND. If the slope of the gold price regression for the next full year is negative with a correlation of <= -0.22 then the prediction is marked as CORRECT and WRONG otherwise with a possible refinement of WRONG TREND CHANGED if the absolute value for the full year correlation was better than 0.22.

Distribution

The pattern of forecasts results produced the following results vector (from 1970, 6 quarter regression ending December 31st of each year and threshold equal to 0.22, with the results from 1990 to 2015 shown in bold):

"T" "T" "T" "T" "F" "T" "F" "T" "T" "F" "F" "F" "NA" "NA" "F" "F" "T" "F" "F" "T" "NA" "T" "F" "F" "T" "NA" "T" "T" "F" "T" "NA" "NA" "T" "T" "T" "T" "T" "F" "T" "T" "T" "T" "NA" "T" "T" "tbd"

Where

  • T - December forecast was correct for the following year

  • F - December forecast was wrong for the following year

  • NA - no forecast was made for the following year (pattern for prediction not matched)

  • tbd - results are to be determined based on 2016 gold prices

Actual correlation by year follows:

Year Correlation
2015 -0.86
2014 -0.63
2013 -0.89
2012 0.24
2011 0.84
2010 0.94
2009 0.86
2008 -0.67
2007 0.84
2006 0.47
2005 0.82
2004 0.56
2003 0.75
2002 0.84
2001 0.70
2000 -0.75
1999 0.03
1998 -0.24
1997 -0.91
1996 -0.91
1995 0.38
Click to enlarge

Prediction Rule Back-test Results

For each year from 1990 to 2014, the per-year plot in this section shows the 6Q regression line computed at the end of each year using the correlation threshold of 0.22. The forecast for the following year does NOT have look ahead bias.

The pale blue line shows the actual regression line for the following year. The details at the top of each plot summarize the regression run parameters used, the resulting regression line (dark blue) slope with correlation/slope and the forecast for the following year (made with zero look ahead bias). The second line in each chart provides a retrospective ranking of the forecast based on the full next year regression line in the following year. The forecast result is one of "CORRECT", "WRONG", "WRONG TREND CHANGED", "NA" and "t.b.d.". "NA" applies when a forecast was not made for the following year and "t.b.d." applies for 2016 since the actual 2016 regression line is unknown.

Note: In all of the back-test runs completed (varying the number of quarters used in the regression and varying the correlation threshold used in the decision rule), the 2016 forecast was always for another down trend year with odds better than even money.

My results are generated using the R programming language (ggplot2 and the linear model regression) and be easily reproduced or extended. My gold price data set used East Coast 4PM spot market prices (from 2012 forward) and LBMA PM prices (or AM prices if the PM price wasn't posted) for earlier dates. The correlation between East Coast 4PM spot market prices and LBMA PM closing prices is 1.0 - see the GoldMinerPulse link provided for the plot.

Results for the 2015 prediction (down trend continues) will only be decided at the end of 2016 (or at least well in the year when the year trend is well established).

2015

Click to enlarge

2014

Click to enlarge

2013

Click to enlarge

2012

Click to enlarge

2011

Click to enlarge

2010

Click to enlarge

2009

Click to enlarge

2008

Click to enlarge

2007

Click to enlarge

2006

Click to enlarge

2005

Click to enlarge

2004

Click to enlarge

2003

Click to enlarge

2002

Click to enlarge

2001

Click to enlarge

2000

Click to enlarge

1999

Click to enlarge

1998

Click to enlarge

1997

Click to enlarge

1996

Click to enlarge

1995

Click to enlarge

1994

Click to enlarge

1993

Click to enlarge

1992

Click to enlarge

1991

Click to enlarge

1990

Click to enlarge

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.