In recent months gold prices have taken quite the beating. Since the 2008 financial crisis gold has been more a speculative investment, with investors worried about inflation and poor economic growth using gold as a safe harbor for their money, many of those investors using ETF's such as (NYSEARCA:GLD) as the medium.
As Tim Lordanov states in a recent analysis on gold prices, gold isn't a currency, it has no intrinsic value, gold demand in China and India may be uncalled for, and gold is probably an asset bubble.
Now you may argue with some of the 50,000 foot level analysis that Tim makes, but I take a bit different approach when looking at gold price trends. I am an amateur economist and statistician (yes amateur, I do hold a degree in economics, but most of the financial pricing that I do is simply hobby and in no way professional), and I enjoy finding variables that I think are heavily tied to the pricing of certain assets.
Using linear regression analysis, I have identified some key variables that I find important to the pricing of gold. The first variable I used in my linear regression model was the consumer price index for the United States. With all of the mentions of gold as a hedge against inflation, it seems obvious that CPI and gold have a positive correlation.
The next variable I used was the University of Michigan Consumer Sentiment Index. As I mentioned in the opening paragraph, gold is an asset that investors tend to use as a safe harbor. When consumer sentiment decreases gold price increases. My model does in fact find a negative correlation between consumer sentiment and gold price.
Next I use the U.S. Ten-Year Treasury rate (index value). It is no secret that gold reacts to monetary policy (for example look at any date that Federal Reserve Chairman Ben Bernanke gives indication about future policy and look at gold prices that day). When the Ten-Year rate increases, gold tends to have downwards pressure, and vice versa.
The final variable I used was the index value of gold imports in the U.S. (adjusted for price change). When gold has higher demand in the U.S., prices should generally increase. I realize that this somewhat ignores the international demand, such as the above mentioned China and India demand, but I think it does give a good general picture of gold demand, although somewhat incomplete.
The results leave me with an R-squared of about 74% (the variables explain about 74% of the price fluctuations) - not the greatest, but definitely a good indication of the majority of variables affecting gold price.
So I made a graph with the predicted gold price by my model vs. the actual gold price. The blue line represents the actual gold price, the red line is the predicted gold price, and the green line is the difference between the blue and red lines. Keep in mind that I used quarterly data.
Now it is quite obvious that linear regressions aren't always right on target, but again this model shows about 3/4's of the factors representing gold pricing.
In the early 2000's my model shows that gold was a bit overpriced, and in fact we see a trend of falling gold prices in the late 90's. Let's focus on the important parts of the graph, when the predicted and actual price of gold cross.
In the year 2000, the lines cross. Looking at the chart below, we see that a significant decrease in the Ten-Year and a decrease in consumer sentiment drives my predicted price higher. At this point the residual (difference between the lines) is lower than $200, or in other words my model predicts a price that is over $200/ounce higher. I consider the residual value of -$200 or under to be a good time to buy and the +$200 mark or above to be a good time to sell.
Using the $200 threshold, the model would have told me to buy from January 2001-January 2002, July 2002-July 2003, April 2005-October2005, and April 2008-January 2009.
The two lines cross again in late 2009, representing a switch from bullish to bearish concerning gold price. The residuals show gold getting quite over priced, and perhaps this is due to the increased speculative behavior mentioned in Tim's article.
The model crosses the +$200 threshold in 2010, representing a suggested sale of gold. At the current date, the model has a residual of $490, still well over the selling threshold.
Let me conclude by pointing out the obvious. The model crossed the selling threshold in late 2010, completely missing the 2011 peak in prices, which would have been a great time to sell. But I keep in mind the fact that this model shows about 75% of the factors, there are other variables as well. If I would have actually traded on this model and bought in the periods mentioned and sold in the periods mentioned, I could have made some pretty decent returns.
My linear regression uses a solid mixture of fundamental analysis mixed with pricing analysis to predict pricing. As always models can be off. I figure my model has performed relatively well, which is why I am confident that the combination of recently increased Ten-Year rates, increased consumer sentiment, slowly increasing CPI, and bearish imports create a clear downward pressure on gold prices.