The classic approach to developing trading systems is to develop a set of trading rules and parameters that work well on one particular market. One of many problems this can create is that the system tests will generate so few trade samples that the system’s hypothetical profitability is sheer coincidence.
Consider a trading system that shows tremendous potential but generates only 100 trades over a significant period of time. It is possible that this great performance is the result of randomness and is meaningless. Let’s look at an example that shows how a seasonal trading approach that advocates always buying on one day of the year and selling on another day of the year may be just as meaningless.
First, we create a grid using random numbers with 15 rows representing years and 12 columns representing months of each year. In each grid, we will randomly generate a 0 or a 1. A 0 represents a down month and a 1 represents an up month. With enough iterations, we will at some point find a combination of numbers that appears statistically significant. In 15 years of randomly generated numbers, we should generate 7.5 up years and 7.5 down years.
In our example, if 0 occurs 13 or more times in a given column with two 1s or fewer, it would be a statistically significant sign that the market is going down. If there are 13, 14 or 15 number 1s in a column, it is statistically significant that the market is going up. When running this experiment, there will be several occurrences that look significant, but we know that the numbers being generated are random; therefore, these results are meaningless.
Just as chance can create a “seasonal” condition that appears statistically significant, chance can create an optimized trading system that also appears significant.
To lessen the likelihood that we mistake unprofitable systems for profitable ones, we can develop our systems on multiple markets using the same rules and parameters. This provides a much larger trade sample than a single market.
The independence of individual trades is related to the markets and time frame you are trading. Shorter-term systems can result in trades that are less correlated over similar markets, such as a collection of stock indexes. If trades last just a few days, the same system and parameters can result in surprisingly low correlations.
This means that a system, using the same rules and parameters, can trade three similar markets — say, the E-mini S&P 500, Nasdaq 100 and Russell 2000 — profit in all three but only increase the maximum drawdown by, say, 20%. Additionally, by trading three markets, the trade sample is potentially three times larger, meaning the results are more statistically significant.
No matter how many markets you trade, drawdowns will not be eliminated. A single trading method or industrial group still has enough correlation that a single system on a set of markets can still have periods of drawdown or flat performance.
In addition to trading a single system on many markets, you can trade multiple systems, as well. Care must be taken, however. Often, strategies are just as, if not more, correlated than markets. Consider most commercially available trend trading strategies. They will be highly correlated to traditional strategies, such as channel breakouts. This is because the big moves that make money tend to be so big that they trigger multiple trade entry rules.
In fact, there is little, if any, benefit because similar systems can show a correlation of 0.70 or greater. Instead, you should trade systems based on different methodologies on baskets or different market sectors.
Consider trading three different systems, We will use each with a $250,000 portfolio for each. One trend-following system will trade corn, wheat, live cattle, cotton, sugar, lumber, palladium, crude oil, natural gas, the U.S. Dollar Index, the Japanese yen, Treasury notes and London copper. We will combine this with a short-term swing system that trades, Nasdaq and Russell 2000 mini futures. Finally, we will trade a third system created for interest rate markets, which trade the 30-year Treasury bonds.
We will run all of them over the period from Nov. 27, 2002 to May 13, 2016. We will use 10% margin so if we trade 10 markets with a $250,000 account, we will allocate $2,500 in margin for each market, which means many of the contracts and trades are skipped until our equity increases.
During this period we produced a compound growth rate of 23.41% with a maximum drawdown of 37.87%. Deducting $37.50 per round turn per contract.
The swing system trading the two stock indexes did not do as well; compound growth rate of 13.11% with a 49.74% max drawdown. The 30-year Treasury bond system using the same $250,000 over this period made 16.91% with a 46.36% maximum drawdown. If these systems were highly correlated we would expect a drawdown of more than 70%. However, when all three systems using the same pool of money were tested, total profits remained while the drawdown increased only marginally over what any of the three systems generated by themselves.
We will allocate $750,000 as starting capital for these three systems and use our 10% margin approach. When we combine these systems we find a compound annual growth rate of 24.98% with a 26.06% maximum drawdown for a margin-at-risk ratio of .9582. By combining the three systems produce better risk adjusted returns.
These results might be unrealistic and for these simple system we would not expect the same out of sample performance but, the results demonstrate that diversifying both across markets and trading systems can ad value.
We can capture the same diversification benefits ourselves, by developing trading systems with this concept in mind.
Our portfolio will be a modified version of the trend following portfolio used earlier. We will trade corn, wheat, live cattle, cotton, sugar, lumber, palladium, crude oil, natural gas, the dollar index, the Japanese yen and T-notes.
For our trading logic, we will use three classic systems, developed on the TradersStudio platform. The first is a triple moving average crossover system. We buy when a shorter-term moving average crosses above a longer-term moving average and sell when the shorter-term average crosses below the longer-term average (see “System 1-3” page 55. Code is written for TradersStudio, but the logic is easily adaptable to other platforms.)
Our test reflects $37.50 deducted for slippage and commissions, and the results are optimized across all seven of our markets. The time frame for the test is November 2002 through May 2016. Results provide a good lesson of how to use optimization. Don’t pick the top set of parameters because the nearest parameter set only produced one-quarter of the profits. The six, 60 and 85 set of parameters produced strong results with similar results in adjacent values. This combination produced $770,115 on an average trade of $11,705 with a maximum intraday drawdown of $111,129.
When examining individual markets in the portfolio, different sets of parameters do better or worse on individual markets. However, our goal is overall stability plus an adequate number of trades for statistical validity. This is important when you have less than 100 trades per market. By optimizing across the basket of markets, there are 325 trades in our sample, which is not a lot, but better than 50.
Our next system is an index trading system based on an opening range breakout. The optimization process for this opening range breakout system will determine which percentage of the average range to use. We optimized the percentage of the three-day average range used from 0.10 to 1.00. This system also has another little trick that Larry Williams discussed in his book, “Long-Term Secrets to Short-Term Trading.” We buy only if the market closed lower than it opened the day before and sell only if it closed higher.
This system was tested on the E-mini Nasdaq and Russell 2000. Each trade will use one mini contract of each. We’ll use a slippage and commission deduction of $60 per contract. In our testing, 90% of the range produced the best results. This system produced $115,575 in profits during the testing period on one mini contract of each. However, the drawdown was $41,470, which is high for this profit level.
Our final system will be used to trade Treasury bonds. The bond system is an intermarket divergence trading method that uses an intermarket trigger, based on the logic that interest rate futures are negatively correlated to the intermarket.
Using $40 per trade for slippage and commissions, we optimized both sets of parameters from two to 40 in steps of two. The test period for this system test is September 1987 through May 2016. This produced 400 parameter combinations. Some combinations lose money. We also have other groupings of parameters that perform well. The parameter set we settled on is six and 20, which produced $270,225 across the three markets with a maximum intraday drawdown of $22,587. There were 499 total trades during the testing period.
Let’s combine these systems to see what effect they have on the return-to-risk ratio. Looking at the systems together, simply adding the drawdowns of $111,129, $41,470 and $22,587, totals $175,186. This would require an account of at least $250,000 just to cover margin and drawdown. More realistically, we would need at least $300,000 to trade.
However, this is only part of the story. To take advantage of the full power of portfolio trading, we need to scale our trades appropriately, which means money management is introduced into this mix. In other words, we size our trades so the risk is normalized across markets. TradersStudio is one platform that allows us to do this automatically.
Our portfolio trades on a 10% margin, so if margin is $4,000, then we require $40,000. Likewise, as the account size grows, so does the number of contracts traded. We will start with $750,000. We also divide the available funds equally between each system session. We then divide by the number of markets at the session level. The results for this portfolio are rather substantial and reflect the amount of money necessary to trade so many markets, as well as the compound growth rate.
The total profits of the combined systems are $16.75 million on a maximum intraday drawdown of under $2 million. There’s an addition $1.4 million in open position profits. The percent profitable is 49%, while the profit factor is 1.52. There are 1,450 trades in the test (see “Compound growth,” page 56).
A major point with this analysis is that even a simple, well-known system such as the moving average crossover can provide real benefits when it is included as part of a portfolio of all systems. One goal for a trader is to have steady returns. Trading a diversified portfolio, with multiple systems that aren’t overly complicated or excessively optimized can get you there. Of course, it also helps to have a well-thought-out and consistent money management plan that will allow your profits to grow as you can afford more exposure to the markets.