Gold, the one asset that has been used for hundred’s of years as a means of currency and wealth accumulation, is used today primarily as a hedge. Since peaking in the late 1980s, gold prices steadily declined, then held around $300 per oz. through the late 1990s. Then starting in late 2001, and during the past six years, gold prices have trended higher, surging to an intra-month peak of $732 in May 2006.
In hindsight, a smart trader would have bought as much gold as possible at or below $300. But that is easier said than done. Catching the beginning of a trend is difficult because a trend must first establish itself before it can be recognized as such. Trends are a result of future expectations. If the expectation is that gold prices will appreciate over the long term, then a new up trend will establish itself. Just to demonstrate that trading isn’t even that easy, it’s equally hard to determine when a trend has ended or if a sudden sell-off is only a temporary correction.
The key is to distinguish the subtle clues that signal possible turning points for gold and adjust accordingly before the move. This, of course, is a Herculean effort. However, with artificial neural network technology, money managers, traders and investment institutions have a better chance to identify those subtle clues and prepare for turning points ahead of the tape.
THEORY AND REALITY
While there may be thousands of smaller catalysts for the rise in gold, we don’t have to identify every one of them. Indeed, as in most intermarket analysis based trading techniques, it’s necessary to look at only a few related markets. For gold, it’s sufficient to look at the U.S. Dollar Index, the Australian dollar and yields on the 10-year Treasury note. Each of these markets has a fundamental link to gold that helped in its selection.
The yield on the 10-year T-note is considered an indicator of U.S. economic health. Changes in the yield affect inflation expectations and the overall forecast of U.S. economic strength. The general market belief is that gold prices are directly correlated with T-note yields: The lower the yield, the greater the risk of inflation and the greater the price appreciation of gold.
The Dollar Index has the wide reputation that it’s inversely correlated with gold prices. A weak U.S. dollar is, almost by definition, indicative of rising inflation, which in turn provides a catalyst for rising gold prices. Conversely, the consensus is a strong dollar should lead to lower gold prices.
The Australian dollar, on the other hand, has more of a brick-and-mortar relationship with gold. Australia is the third largest gold-producing country in the world. Unlike the U.S. dollar, it’s widely believed that the Australian dollar is directly correlated with gold prices; as the Australian dollar appreciates, so should gold.
After the logistical step of collecting the data, the next step is to perform some preliminary analysis. Two types of statistical measures reflect the robustness of the inputs. These are the coefficient of determination (R^2) for all three input variables in relation to the output variable and the individual t-statistics for each input variable (see “Robust or bust,” right).
The coefficient of determination helps us understand the total variance of our output variable in terms of the input variables, and the t-statistic helps us understand the statistical significance of each input variable to the output variable. A t-statistic value of more than 2 or less than -2 for each input variable indicates a statistical significance.
Interpreting the table, we can say that all three of our input variables explain about 74.9% of gold’s daily price fluctuation, and the three input variables are statistically significant.
FINDING TURNING POINTS
Although it’s nice to know that our selected fundamentals are statistically significant, we need more horsepower than simple statistics provide to use them to forecast trends in gold prices. This is where artificial neural networks come into play.
The power of these tools allows us to analyze reams of financial data for subtle relationships between input and output variables. Once those relationships are determined and a model created, we can graph the various relationships and visually identify important price levels. In some cases, we discover relationships that confirm our market beliefs. In other cases, we denounced them.
The first test is to confirm the direct correlation between the Australian dollar and gold. The neural network model indicates a nearly linear relationship between the currency and gold prices (see “Almost linear,” above). The greater the appreciation in the currency, the greater the appreciation of gold. (Note that to highlight the effect of a specific variable, the other variables must be held constant. To construct this graph using the artificial neural network, the Dollar Index was set to 82 and the T-note yield to 5.12%. The Australian dollar varied in a range of 0.636 to 0.840.)
The next test examines whether the Dollar Index behaves inversely to gold prices. The result, instead, is an interesting non-linear, bell-shaped relationship (see “On the curve,” above). Gold prices tend to peak when the Dollar Index is at 89 and then decline as the index declines.
As the dollar gains strength and rises, gold prices will act inversely and fall. (For this graph, the Australian dollar was held at 0.8340, the T-note yield at 5.12% and the Dollar Index varied from 99.1 to 80.6.)
Finally, we looked at whether the 10-year T-note yield is directly correlated to gold prices. This test provided another surprise as it also identified a nonlinear relationship between the variables (see “Off the edge,” below). Yields and gold prices are directly correlated until the yield peaks at 4.8%; thereafter, gold prices fall as the yield increases. (This graph held the Australian dollar at 0.8340, the Dollar Index at 82 and varied T-note yields from 3.75% to 5.25%.)
This type of analysis is a powerful tool for anyone involved in the gold futures market. One helpful application is to determine the outcomes of various “what-if” scenarios. To illustrate, let’s take a “what if” scenario, apply the model, determine predicted outcomes, and then compare it to a real world event.
For illustration, assume that you review your positions one day and note that the Australian dollar has been trading in a range between 0.8200 and 0.8350 for several weeks, the Dollar Index is range-bound between 81.00 and 82.50, and the T-note yield is near 4.71%.
Suddenly the news flashes across your trading screen that the bond market starts to sell off following the release of some data that suggests the U.S. economy is overheating. Bond traders expect the Federal Reserve to raise interest rates at their next meeting to rein in the growth. This news spills over into the currency market as the Australian Bank issues hawkish statements regarding that country’s inflation and the Australian dollar moves higher. The Dollar Index, by comparison, barely moves.
What is the implication for gold based on what we’ve learned? Applying the model to this scenario suggests that gold prices are expected to peak as bond yields increase from 4.71% to 4.85% (see “Range of scenarios,” above), and then decline rapidly as yields approach the 5.1% area. Roughly, we can expect that a 7.6% increase in bond yields will translate to a 6.2% decline in gold prices (see “Where gold should go,” above). This last gasp price appreciation for gold prices between the 4.71% and 4.85% yield area should present a nice opportunity to take profits or close positions.
A similar event to this scenario actually happened between June 1 and June 14, 2007 (see “Reality unfolds,” below). On June 1, gold prices closed at $671.20, the 10-year T-note yield was at 4.86%, the Australian dollar was at 0.8229, and the Dollar Index closed at 82.18.
In the span of 10 trading sessions, gold prices tumbled by 2.92% as bond yields increased by 5.35%. During this same time frame, the Australian dollar rallied past 0.8300 and the Dollar Index remained in a range between 81.66 and 82.31. What’s interesting to note is that gold prices actually peaked at $671.20 on June 1 after closing at $661 on May 31. Bond yields for both days were at the inflection point on the non-linear relationship curve as shown on “On the curve.”
The major difference between the illustrative what-if scenario and the actual event was that the Dollar Index slightly increased and the Australian dollar rallied more than in the scenario. Just these two slight differences help lessen the decline of gold prices from a predicted decline of 6.2% to the actual decline of 2.9%.
One thing that’s clear from this exercise is that gold prices are heavily influenced by Treasury yields. This is corroborated by the strong statistical significance of the yields. We also learned that while bond yields greater than 4.85% will negatively impact gold prices, a strong Australian dollar can act as temporary support, thereby lessening price declines. In the event that the Australian dollar weakens, we should expect gold prices to decline rapidly if faced with persistently high bond yields. Conversely, if bond yields are less than 4.85% and the Australian dollar remains strong, it would be reasonable to expect gold prices to rally.
Special thanks to Stephan Kudyba for his contributions to this article.
Thomas N. Ott is an associate for Booz Allen Hamilton and can be reached at email@example.com. He maintains a web log dedicated to modeling market trends using neural networks at www.neuralmarkettrends.com.