Mining For Maximum Shine

"One big difference between commodities [futures] and stocks is that the former have very definite rhythms," points out Darrell Jobman, editor-in-chief of TradingEducation.com. "Unlike stocks, futures do not continue moving higher for years on end, and they rarely if ever go to zero."

Futures traders have long employed cycles to time buys and sells based on periodic demand and supply paradigms, which can vary greatly. Natural product futures such as grains, meats, and softs (lumber, cocoa, and cotton, for example) have definite growing and harvest seasons that affect prices. The same is true when it comes to exploration and recovery of energies and minerals.

Over the years, traders have learned that tracking crop and weather reports are helpful in choosing times to buy and sell, and seasons play a big part in this process. When it comes to metals, seasons are also important for two reasons. First, exploration and recovery is often highly weather dependent. So are periods of high demand (as in, say, Christmas).

Interestingly, however, precious metals such as gold and silver have different supply-demand cycles so it is necessary to know what drives the underlying fundamentals before trading them. Assuming any two are alike can be a costly mistake. Learning these fundamentals takes time -- lots of it.

Surely, in this day and age of high-speed computers, there must be a better way.

A number of the widely used computer charting programs allow traders to test for optimal buy and sell times. But test the data over different periods and you will find this has changed over time. Employ five years of data (2000-05) and the graph looks different with a low in late April followed by another low in late July after which gold prices move up to the end of the year (see Figure 1). Figures 2 and 3 show weekly charts of silver and palladium.

FIGURE 1: WEEKLY CHART OF ENGELHARD GOLD PRICES. Here you see the seasonal July buy and December sell signals. The equity curve is in the upper supbgraph. METASTOCK; DATA FROM WWW.PRING.COM

FIGURE 2: WEEKLY CHART OF SILVER (1995-2005). A seasonal August buy and March sell signals for silver produced the equity curve at the top. METASTOCK; DATA FROM WWW.PRING.COM

FIGURE 3: WEEKLY CHART OF PALLADIUM. A November-March seasonal on the weekly chart of palladium resulted in the equity curve in the upper subgraph. METASTOCK; DATA FROM WWW.PRING.COM

-----------------------------------------------------------------------
Buy when Month = 7 and Day of the Month is equal to or greater than 23.

Sell when Month = 12 and Day of the Month is equal to or greater than 23.

(Hint: When using weekly data, it is important to widen date parameters to ensure the program has enough data to find the last week of the month, hence day of the month = or > 23.)
-----------------------------------------------------------------------

This formula will work well to determine the returns over various periods by buying and selling at specific times of the year.

You could spend hours poring over stacks of commodity books to find the answer. Even if you did, there would be no guarantee the information you'd glean would be correct, since season patterns have a habit of changing over time.

This is where a computer, a charting program, and some formula language knowledge can be a lot of help. Using our logic above, here is the formula for finding the optimal buying and selling periods:

-----------------------------------------------------------------------
Buy if Month = Optimization1 and Day of the Month is equal to or greater than 23.

Sell if Month = Optimization 2 and Day of the Month is equal to or greater than 23.
-----------------------------------------------------------------------

This tells your computer to test for the best buying and selling times over a given period. The test requires 144 runs (12 months for buys multiplied by 12 months for sells), which should take the average computer today less than a minute. However, this assumes you are going to stick with buying and selling during the last week of the month. If you also want to test for the best day of the month, it requires two more variables (optimization periods). The formula looks like this:

-----------------------------------------------------------------------
Buy if Month = Optimization 1 and Day of the Month = Optimization3.

Sell if Month = Optimization 2 and Day of the Month = Optimization4.
-----------------------------------------------------------------------

Assuming you test every other day, this requires the computer to run 32,400 tests, which will take considerably longer. Another option is to run two separate tests, the first to find the best months to buy and sell using the last week, then rerunning the tests to find the best days of the month.

Figure 4 shows the results of our tests on weekly metals data. The test period covered a total of 573 weeks or 4,011 days between January 4, 1995, and December 30, 2005.

FIGURE 4: PERFORMANCE RESULTS. Here you see the results for the most profitable periods to buy and sell various metals using data from 1/4/95 - 12/30/2005. CREDIT: WWW.TRADINGEDUCATION.COM

This is why it's important, once you have completed your tests over various periods and developed what you think will be a killer new trading program, to test it first in out-of-sample data, completely separate from the periods used in your tests. If the system is good, results should hold up. However, more often than not, you will be dismayed that the results fall dramatically.

Assuming your system survives rigorous out-of-sample tests, the next step is to test it in real time using a simulated account. Many brokers now offer this service to allow traders to test their acumen before risking real money. A number of these systems are better than paper-trading because they utilize real data and fill conditions to execute simulated trades. This will avoid cases of curve-fitting (designing your system to work over the test period by overoptimization). In most cases, you will find results will vary, depending over what period you test your cycle system.

This process for developing a trading system is not restricted to metals. It works for testing any futures, currency, bond, or stock. If backtesting over different periods fails to generate consistent results, it means either there is no tradable cycle or the cycle changes often. Discard the subject of the test and move on.

However, if you determine a market has seasonal lows in, say, July in the case of gold but is putting in new highs when the seasonal time to buy comes around, what do you do then? That will depend on what other technicals are saying. Is gold in an uptrend with trendline support? Buying the next time it hits support and setting stops just below the trendline is the course of action. If gold breaks support, you are out for a small loss, but if you have caught the season correctly and bought on weakness, then you stand to profit. All good systems should be used in conjunction with confirming technicals to reduce the chance of failure and account meltdown.

Cycles are one more technical tool to help you decide the best times to buy and sell your metals. With a little luck, some computer know-how, and a lot of hard work, you even stand a good chance of making money!

Matt Blackman, market analyst for www.TradingEducation.com , a free educational website, is a technical trader, author, reviewer, and keynote speaker whose work has appeared in a number of major financial publications, websites, and newsletters. He is a member of the Market Technicians Association (MTA) and Technical Securities Analysts Association (TSAA).

SUGGESTED READING
Moore, Steve, Jerry Toepke, and Nick Colley [2006]. The Encyclopedia Of Commodity And Financial Spreads, John Wiley & Sons.

Upperman, Floyd [2006]. Commitments Of Traders: Strategies For Tracking The Market And Trading Profitably, John Wiley & Sons

Williams, Larry [2006]. Trade Stocks & Commodities With the Insiders -- Secrets Of The COT Report,John Wiley & Sons.

Seasonality in Industrial Commodities and Equity Indexes http://dvtechtalk.com/specialreports/specialreport7.htm