DERIVATIVES Trading Soybean Spreads Here are the basics of trading a soybean commodity spread using a seasonal strategy. by Scott W. Barrie T he price relationship between two or more given commodity contracts is known as a spread. Spread trading is the purchase of one commodity contract and the simultaneous sale of another, related, futures contract. The price difference can change, and if it trends in the correct direction, the change in the relationship of the prices will be profitable. There are two basic types of spreads: intercommodity and intracommodity spreads. An intercommodity spread is the purchase of a given commodity and the simultaneous sale of another related but different commodity. Examples of common intercommodity spreads are: the Treasury notes Treasury bond spread, called the NOB spread; the corn wheat spread; the T-bill Eurodollar spread, called the TED spread, and the live cattle feeder cattle spread. Trading intercommodity spreads involves speculation on the relationship between related but different markets. An intracommodity spread involves the purchase of one delivery month and the simultaneous sale of a different delivery month of the same commodity on the same exchange. A common example of an intracommodity or interdelivery spread is the July November soybean spread. THE PROS AND CONS Since spread trading involves the simultaneous purchase and sale of two (or more) commodity contracts, the trading of spreads involves a higher initial overhead cost in the form of commissions. Although most firms will give commission discounts for spread trading, the discounted commissions are typi- HOLLIS BOGDANNFY
cally still higher than the cost involved in a simple outright position. Spread trading, though generally accepted as less risky, can entail more risk because you have the opportunity for two positions to move against you, as the price of the long position contract can go down and the price of the short position contract can rise. As such, individuals should closely examine the fees and risks involved in spread positions before trading. Spread trading does have several distinct advantages that outweigh the added costs associated with this type of trading: The spreads have attractive margin requirements. The margin requirement is the funds you deposit with your broker. For outright positions there is one level, and for spread trading the capital required is generally half or less. There is less risk because the day-today changes in the price of the spread is typically less than the day-to-day change in price of an outright position. The price of the spread can demonstrate increased predictability due to the seasonal nature of the spread markets. For these reasons, spread positions are considered hedged positions. The long position hedges the short position, because if the long side of the spread is showing a loss, the short side of the spread should be showing a profit, or vice versa. This is the case most of the time. The hedged nature of spread trading can also be seen in the drastically lower margin requirements for spreads. In addition, because spreads are relationships, they tend to behave in a more rational manner compared to outright positions. THE OLD CROP NEW CROP SOYBEAN SPREAD The most commonly followed soybean interdelivery spread is the July November spread (Figure 1). This spread is a classic example of an old crop (July) to new crop (November) spread. Soybeans are typically planted in March and harvested in November; as such, soybeans for July delivery are old crop, as they have been sitting in storage since last year s crop, while the November contract is this year s crop. This spread is in essence a measure on the immediate demand for soybeans, as opposed to the future demand for soybeans. FIGURE 1: SPREAD MARKETS. The top chart is the July 1996 soybean contract and the middle chart is the November 1996 soybean contract. The bottom chart is the spread chart, which is the difference between the closing prices of the two contracts. Notice that the spread chart has an uptrend, a downtrend and sideways trading patterns. When the immediate demand for soybeans is high, the July contract will increase in value relative to the November contract. Thus, if prices rise, then July soybeans will increase in value by a greater amount than November soybeans; if Soybeans are typically planted in March and harvested in November; as such, soybeans for July delivery are old crop, as they have been sitting in storage since last year s crop, while the November contract is this year s crop. prices decline, then July soybeans will decrease in value less than November soybeans. When the immediate demand for soybeans is low, the July contract will decrease in value relative to the November contract. This means that if prices rise, then July soybeans DECISION LOGIC FILTER USING THE HIGHER WEEKLY CLOSE OR LOWER WEEKLY CLOSE System Week# Parameter Action Holding # of # Wins # Loss Total Avg Avg period trades P&L profit loss SS-1 51 HC Sell 2 15 13 2 106 7/8 8 5/8-2 SS-2 1 HC Sell 3 11 10 1 68 1/8 7 1/4-3 SS-3 2 LC Sell 2 13 11 2 162 7/8 15-1 SS-4 21 LC Sell 4 15 12 3 212 23 3/8-23 SS-5 23 LC Sell 3 16 13 3 164 1/4 13 2/4-3 SS-6 24 HC Sell 2 15 13 2 138 1/2 11 3/4-1 FIGURE 2: LOGIC FILTER RESULTS. Here are the results for each system. The specific rules are listed in the sidebar titled Rules for decision logic-based seasonal trading. TT CHARTBOOK (TECHNICAL TOOLS)
will increase by a lesser amount than November soybeans; if prices decline, then July soybeans will decrease in value more than November soybeans. BUILDING A SYSTEM Like all agricultural markets to some extent, the soybean market is seasonal in nature; as such, I wanted to incorporate this seasonality into my trading system. This system is a robust one for trading the July November soybean spread that incorporates the seasonal nature of both the soybean market and the soybean spread markets using event-based logic. Toward this end, the following controls are used for development and testing. The period covered is November 1 through July 20, 1969, through 1996. The spreads are quoted as the July delivery contract price minus the November delivery contract price, within the same calendar year. Long positions entails buying July and selling November; short positions indicate selling July and buying November. Therefore, the spread is always quoted as the position of the July contract. This system is based on two parameter sets, the number of the calendar week and the close of the current week in relation to last week s data. The first Friday (or last trading day of a calendar week) of the new year is always considered week 1, while the last Friday of the year (or last trading day of the last full week) is always considered week 52. Week numbers are assigned sequentially from these points. DECISION LOGIC FILTERING The basis of seasonal analysis is that the current trading time frame will act normally meaning that this year will behave as the past years have, and hopefully the most profitable past years. In order to isolate the most profitable years from the dataset, I used an event that must be triggered in order to have a complete setup for the trade. The decision logic being used for this example is a simple pattern: either a higher weekly close or a lower weekly close, indicated with an HC or an LC, respectively, in Figure 2 (see sidebars Rules for decision logic based on seasonal trading and Performance breakdown of the decision logic seasonal trades ). The week column denotes the week number in a standard 52-week year, as denoted in the controls for testing. The parameter is the decision logic to take the trade, with LC signifying a lower close on the appropriate week number as compared with the previous week, while HC signifies a close higher than the previous week s close. The action column is the appropriate trade to be taken, with sell meaning a short HYPOTHETICAL SYSTEM PERFORMANCE ON YEAR-BY-YEAR BASIS Year # of # of # of Total profit/ trades winners losers loss ($) 1969 3 3 0 $187.50 1970 2 2 0 562.50 1971 3 3 0 325.00 1972 5 3 2 268.75 1973 2 2 0 4,175.00 1974 2 1 1 0 1975 5 4 1 2,750.00 1976 2 2 0 375.00 1977 3 3 0 10,725.00 1978 3 2 1 487.50 1979 2 1 1 175.00 1980 4 4 0 937.50 1981 2 2 0 2,525.00 1982 5 4 1 725.00 1983 5 5 0 2,212.50 1984 3 3 0 10,362.50 1985 1 1 0 287.50 1986 2 1 1 (12.50) 1987 2 2 0 237.50 1988 4 3 1 (612.50) 1989 2 1 1 2,525.00 1990 2 2 0 737.50 1991 4 2 2 (25.00) 1992 5 5 0 637.50 1993 3 3 0 400.00 1994 2 2 0 475.00 1995 2 2 0 287.50 1996 5 4 1 900.00 FIGURE 3: PERFORMANCE. This table lists the results year by year. SUMMARY OF THE HYPOTHETICAL SYSTEM PERFORMANCE ON YEAR-BY-YEAR BASIS No commission $50/contract round-turn commission # of years 28 28 # wins 24 21 # losses 4 7 % winning years 86% 75% Total profit $42,631.25 $34,131.25 Average year $1,522.54 $1,218.97 FIGURE 4: SUMMARY. The first column excludes commissions and the second includes commissions. position in the July contract and a long position in the November contract. All profit and loss figures are quoted in cents, where one cent in the soybean market is equivalent to $50.00. No commissions or slippage were applied to the results. Figure 3 presents the hypothetical performance of the system on a year-by-year basis. Figure 4 summarizes the performance. BREAKDOWN OF TWO PATTERNS In the interest of brevity, I will detail only two of the six patterns, SS-3 and SS-4. I chose these two because one should have been completed by the time this article is published,
while the other could be viewed in real time. SS-3 is a simple seasonal pattern combination. The rules for entry are simple. If the closing value of the July November soybean spread (July contract minus November contract) for the second week of the year is less than the closing value of this spread on the first week of the year, then sell this spread (establish a short position in the July contract and a long position in the November contract). The tested holding period is for two complete calendar weeks, or until the end of the fourth week of the year. Figure 5 is a year-by-year listing of the trade results based on the seasonal approach. The rows in bold denote years in which trades would have been executed based on the decision logic listed separately in Figure 6. Figure 7 compares the seasonal trade performance to the system with the decision filter. Using the decision logic of a lower weekly close, coupled with the seasonal nature of this spread, gathers the most profitable trades and reduces the drawdowns associated with the seasonal phenomena. Using a simple filter or decision criteria of a lower weekly close on the second week of the year retains 69% of the total profit associated with the seasonal phenomena with less than half the number of trades. The profit to potential loss ratio ( Average profit/maximum draw on profitable trade ) is an attractive 1.22. The maximum drawdown on a profitable trade is an excellent indication of where to place a stoploss for real-time trading of tested systems. This system calls for a stop-loss placed 10 cents above the closing value of the second week, or roughly $525.00 above the entry price. By contrast, SS-4 is based on the seasonality of the July November soybean spread during the latter part of May or, more precisely, the 21st week of the year. This is typically the height of the weather-controlled markets in the soybean RULES FOR DECISION LOGIC-BASED SEASONAL TRADING Here, the rules including the time period for entry and the appropriate action for each spread trade are detailed. SS-1 Seasonal component: 51 st week of the year through the 1 st week of the following year Current year s dates: December 28, 1996, through January 4, 1997 Decision logic: Higher close on the 51 st week (12/28) than the 50 th week (12/21) If decision logic is true, sell July 97 soybeans and buy November 97 soybeans SS-2 Seasonal component: First week of the year through the 4th week of the year Current year s dates: January 4, 1997, through January 25, 1997 Decision logic: Higher close on the 1 st week (01/04) than the 52 nd week (12/31) SS-3 Seasonal component: 2 nd week of the year through the 4 th week of the year Current year s dates: January 11, 1997, through January 25, 1997 Decision logic: Lower close on the 2 nd week (01/11) than the 1 st week (01/04) SS-4 Seasonal component: 21 st week of the year through the 25 th week of the year Current year s dates: May 24, 1997, through June 21, 1997 Decision logic: Lower close on the 21 st week (05/24) than the 20 th week (05/17) SS-5 Seasonal component: 22 nd week of the year through the 25 th week of the year Current year s dates: May 31, 1997, through June 21, 1997 Decision logic: Lower close on the 22 nd week (05/31) than the 21 st week (05/24) SS-6 Seasonal component: 24 th week of the year through the 26 th week of the year Current year s dates: June 14, 1997, through June 28, 1997 Decision logic: Lower close on the 24 th week (06/14) than the 23 rd week (06/07) PERFORMANCE BREAKDOWN OF THE DECISION LOGIC SEASONAL TRADES This table lists the results of each spread trading system. SS-1 SS-2 SS-3 SS-4 SS-5 SS-6 # of trades 15 11 13 15 16 15 # of profits 13 10 11 12 13 13 # of losses 2 1 2 3 3 2 % Profitable 87% 91% 85% 80% 81% 87% Total P&L $5,343.75 $3,406.25 $8,143.75 $10,600.00 $8,212.50 $6,925.00 Average P&L $356.25 $309.66 $626.44 $706.67 $513.28 $461.67 Average profit $429.33 $358.13 $755.68 $1,170.83 $672.12 $593.18 Average loss -$118.75 -$175.00 -$84.38 -$1,150.00 -$175.00 -$43.75 Average draw -$56.25 -$90.91 -$38.94 -$375.00 -$636.33 -$275.00 Average draw on a profitable trade -$4.81 $5.00 -$19.89 -$113.54 -$554.33 -$238.46 Maximum draw on a profitable trade -$275.00 -$87.50 -$512.50 -$687.50 -$4,900.00 -$1,675.00 Χ 2 6.67 5.82 4.92 4.27 5.06 6.67 Stop level -5 3/4-2 -10 2/4-14 -98 1/4-33 3/4 market; at this point, any lack of rain that has been forecast causes soybeans prices to climb, while actual precipitation generally leads to violent price drops. The July November soybean spread, however, acts in a
SEASONAL TRADES BASED SOLELY ON ENTRY ON THE 2 ND WEEK OF YEAR AND EXIT ON THE 4 TH WEEK OF THE YEAR Seasonal Entry Exit P&L Drawdown Draw on trades price price in cents in cents profitable trades in cents 1969 23 1/8 26 3/8-3 1/4-3 1/4 1970 15 3/8 11 7/8 3 2/4 2/4 2/4 1971 25 24 2/4 1/2-1 1/8-1 1/8 1972 17 5/8 20 2/4-2 7/8-5 1/4 1973 63 1/4 66 5/8-3 3/8-9 1974 15 2/4 16-2/4-2/4 1975 42 2/4 21 2/4 21 2/4 2/4 1976-13 2/4-13 - 2/4-3 1977 46 3/4 36 10 3/4-10 1/4-10 1/4 1978 27 1/4 13 1/4 14 2 2 1979 42 1/4 41 2/4 3/4-3 1/4-3 1/4 1980-23 2/4-27 3 2/4-1 1/4-1 1/4 1981 18-30 2/4 48 2/4 7 2/4 7 2/4 1982-9 -13 1/4 4 1/4-2 3/4-2 3/4 1983-1/4-13 2/4 13 1/4 2 3/4 2 3/4 1984 86 2/4 45 41 2/4 3 2/4 3 2/4 1985 3 2/4 5 3/4-2 1/4-5 1/4 1986 25 1/4 24 1/4 1-16 2/4-16 2/4 1987 15 1/4 15 1/4-1 2/4-1 2/4 1988 3/4-5 5 3/4-4 2/4-4 2/4 1989 89 3/4 39 1/4 50 2/4 11 11 1990-10 2/4-9 2/4-1 -3 2/4 1991-9 2/4-11 1 2/4-5 3/4-5 3/4 1992-9 3/4-13 1/4 3 2/4-3/4-3/4 1993-3 3/4-6 2/4 2 3/4 1/4 1/4 1994 59 1/4 43 1/4 16 2/4 2/4 1995-14 1/4-13 1/4-1 -1 3/4 1996 47 1/4 40 2/4 6 3/4-1 -1 FIGURE 5: SEASONAL TRADES. This table lists every year s performance, and the rows in bold represent the years that decision logic SS-33 was applicable. much more predictable manner during these chaotic times. For the last 28 years (1969 to 1996), the spread has narrowed for 20 of those years between the 21st week of the year and the 25th week of the year. Trading this seasonal bias alone, though, is precarious, as the July November spread has had some violent rallies during this time frame; specifically, the drought of 1973 caused this spread to widen over $1.00 a bushel during our seasonal window. The years 1977, 1988 and 1989 all saw the spread gain over $0.20 a bushel during our seasonal window. But by using the decision logic filter of a lower weekly close on the 21st week of the year, the only year with a rally over $0.20 a bushel remaining was 1988. Using the maximum drawdown on a profitable trade as the guideline for setting a stop on this trade, one could place a stop-loss on this trade as close as $0.14 a bushel on the spread. Figure 8 illustrates the vast improvement upon the seasonal strategy that using decision logic adds. A simple filter of a lower close almost doubles the gross profit of this system while reducing the number of trades by almost 50%. Though the reduced number of trades does degrade the significance of the results, the results are still significant by a chi-square of 4.27 compared with 4.32 of the SS-3 TRADES Lower weekly closes on the 2 nd week, exit on the 4 th week SS-3 Entry Exit P&L Drawdown Draw on price price in cents in cents profitable trades in cents 1971 25 24 2/4 2/4-1 1/8-1 1/8 1972 17 5/8 20 2/4-2 7/8-5 1/4 1974 15 2/4 16-2/4-2/4 1975 42 2/4 21 2/4 21 2/4 2/4 1977 46 3/4 36 10 3/4-10 1/4-10 1/4 1981 18-30 2/4 48 2/4 7 2/4 7 2/4 1982-9 -13 1/4 4 1/4-2 3/4-2 3/4 1983-1/4-13 2/4 13 1/4 2 3/4 2 3/4 1988 3/4-5 5 3/4-4 2/4-4 2/4 1989 89 3/4 39 1/4 50 2/4 11 11 1991-9 2/4-11 1 2/4-5 3/4-5 3/4 1992-9 3/4-13 1/4 3 2/4-3/4-3/4 1996 47 1/4 40 2/4 6 3/4-1 -1 FIGURE 6: This table shows the boldfaced trades from Figure 5. straight seasonal. The average trade profit and loss number is increased by more than $500 because the chaotic activity of 1973 was filtered out. In addition, several zero to two-cent trades were removed, as well as five losing trades. Using the decision logic filter of a lower close turned a marginally attractive seasonal bias into a robust trading system with excellent risk-to-reward characteristics. AND IN CONCLUSION Spread trading requires more patience and a greater understanding of the risks involved than outright position trading does, but the rewards far outweigh the added study. Spreads tend to trend more as well as behave in a more seasonal manner than the underlying futures markets. Those without trading patience, or who have a need for quick and frequent excitement, should probably not trade spreads, but those of you who wish to develop longer-term, accurate trading systems should pay closer attention to the spread markets. By combining simple patterns with seasonal analysis of the spread markets, a decision logic filter can be used to gauge the seasonality of the spread, perhaps take advantage of the most profitable circumstances and avoid the marginal to low percentage profitable trades. Pattern recognition alone is a powerful tool in technical analysis, and combining it with the seasonal nature of the markets can lead to highly robust and historically attractive trading systems. By trading spreads, one is able to trade more conservatively using sound money management because of the lower margin requirements. It is possible for a trader with a $5,000 account to risk 10% of his account while trading a longerterm system and having the stop-loss placed far enough away to avoid the market s noise. Since spread data must be created by subtracting one price series from another, these markets have not been as mined for data as have the underlying futures, so some of the market anomalies have not been as exploited. Spread trading is an underexploited area of the futures
market, one that offers excellent risk-toreward characteristics and a high degree of predictability. Spread markets are not subject to prices trading through levels that trigger stop-loss orders, only to see the prices return to the previous level. Spreads are an excellent tool for taking advantage of the seasonal and longer-term trends of the futures markets. The benefits of trading spreads far outweigh the added work necessary for analysis and the higher costs associated with spread trading. Scott Barrie is the head of research for Great Pacific Trading Co. He edits Great Pacific s Trend Watch newsletter and Seasonal Stratagems Report. Mark Sanders, vice president of Great Pacific, contributed to this report. COMPARISON OF SEASONAL-BASED TRADES WITH AND WITHOUT THE DECISION LOGIC CRITERIA Seasonal Decision logic trades trades # of trades 28 13 # of profits 20 11 # of losses 8 2 % profitable 71% 85% Total P&L $11,737.50 $8,143.75 Average P&L $419.20 $626.44 Average profit $623.75 $755.68 Average loss -$92.19 -$84.38 Average draw -$92.19 -$38.94 Average draw on a profitable trade -$50.31 -$19.89 Maximum draw on a profitable trade -$825.00 -$512.50 Χ 2 4.32 4.92 FIGURE 7: SUMMARY. This table summarizes the results from Figures 5 and 6. SEASONAL PERFORMANCE AND SS-4 PERFORMANCE FOR THE SEASONAL WINDOW OF THE 21 ST THROUGH 25 TH WEEK Seasonal Decision logic trades trades # of trades 28 15 # of profits 20 12 # of losses 8 3 % profitable 71% 80% Total P&L $5,331.25 $10,600.00 Average P&L $190.40 $706.67 Average profit $916.56 $1,170.83 Average loss -$1,625.00 -$1,150.00 Average draw -$657.81 -$375.00 Average draw on a profitable trade -$75.31 -$113.54 Maximum draw on a profitable trade -$687.50 -$687.50 Χ 2 4.32 4.27 FIGURE 8: SS-4 TRADES. The first column shows what the results would be if the seasonal trade had been executed every year. The second column uses the SS-4 decision filter. RELATED READING AND RESOURCE Barrie, Scott [1996]. Pork bellies and the COT index, Technical Analysis of STOCKS & COMMODITIES, Volume 14: October. See Traders Glossary for definition S&C