Calendar Spreads, Outright Futures Positions, and Risk


Publication: The Journal of Alternative Investments

Co-Authors: Ira Kawaller, Paul Koch, and Ludan Liu

PDF Version

A futures calendar spread is constructed by simultaneously buying and selling two futures contracts with a common underlying instrument but different expiration dates—for instance, buying a December S&P 500 futures contract and selling a September S&P 500 contract. While spreads are generally considered to be less risky than outright futures positions, it is important to recognize that market participants typically trade a larger number of spreads than outrights. Presumably, such traders are attempting to achieve greater returns with similar risk, or similar returns with less risk. Depending on the relative sizes of the positions and the performance of the spreads vis-a` -vis the outright, the goal of achieving similar returns or risk may or may not be realized.


These considerations raise several issues. For example, do these different trading approaches display similar performance (i.e., high or low correlations)? What are the relative risks of a calendar spread versus a single outright futures position? Are these risk characteristics stable over time? Given the relative risks, what are the appropriate capital (margin) requirements for trading spreads versus outright futures? Similarly, how many spreads should be traded to achieve comparable risk to a single outright futures position? Finally, what is the relative historical return performance of calendar spreads and outright futures positions with comparable risk?


We propose one way to determine the “appropriate” number of calendar spreads to trade relative to (or instead of) a single outright is to equalize the values-at-risk (VaR) of the two positions. If we assume that daily price changes in a single outright and a single calendar spread are both normally distributed, then the ratio of their respective standard deviations represents the multiple of spreads that offers a comparable VaR to a single outright, ex ante. We call this multiple the ex ante VaR-adjusted spread position. The ratio of standard deviations is appropriate to determine this multiple, however, only if the underlying distributions are normal. Because this assumption may not be valid, it is not clear which strategy would generate better ex post performance in terms of risk and return, or which would experience a return distribution with more desirable properties.


This study investigates these issues in a three-stage analysis, using daily data from 1991- 1997 for 10 of the most active futures contracts traded in the U.S. The first-stage analysis compares the empirical distributions of daily price changes in a single outright futures contract versus a single calendar spread. We test the intertemporal stability of the standard deviations of these two respective positions, as well as the ratio of their standard deviations. This analysis sheds light on the relative return and risk of a single outright futures position versus a calendar spread, as well as possible difficulties in maintaining a VaR-adjusted spread position over time.


In the second stage, we construct the VaR-adjusted spread position on a daily basis, and compare its ex post performance to that of a single outright. If this VaR-adjustment truly equalizes risk, then the VaR-adjusted spread and the single outright position should presumably experience similar return performance over time. In making this comparison we focus on the mean, standard deviation, skewness, and kurtosis of each trading strategy, and we test both positions for departures from normality. The objective is to identify possible systematic differences between these ex post distributions that might be exploited. It is conceivable, for example, that the VaR-adjusted spread could display an ex post distribution with similar average returns, but with a thinner left tail (less downward skewness or smaller kurtosis) than the single outright position. Such an outcome would imply that the VaR-adjusted spread experiences less risk of great loss than the outright.1 This analysis reveals whether the VaR-adjusted spread strategy in fact experiences similar return and/or risk to a single outright position, and illuminates any systematic discrepancies between the two approaches.


Third, we conduct simulations of a trading rule that stipulates buying or selling a single outright futures or a VaR-adjusted spread position based on a moving average of past prices. We then compare the relative performances these trading strategies would have generated over the period 1995-1997. This final analysis sheds light on the relative risk and return offered by ex ante trading strategies involving a single outright futures or a VaR-adjusted calendar spread position.


When a trading strategy that operates on outright futures is compared to one that operates on calendar spreads, trade-offs pertaining to transactions costs versus capital requirements must be considered. Given that many more contracts are required by the spread trader in order to generate a similar dollar-return or bear comparable VaR to an outright futures trader, we might expect transactions costs for trading VaR-adjusted spreads to be considerably higher than those associated with trading a single outright.2 On the other hand, performance bond requirements for spread positions are often considerably smaller than for outright futures positions. This relative advantage of trading spreads is documented in Exhibit 1, which presents the initial margin requirements for trading a single outright futures versus a single calendar spread, for all 10 contracts investigated. The ratio of these two margin requirements is provided in the last column in Exhibit 1, and represents the multiple of spreads that can be traded in each commodity with the same capital commitment as a single outright futures. It remains to be seen whether these relative advantages/disadvantages in terms of transactions costs and capital requirements are maintained after the values-at-risk are equalized for the two strategies. Our analysis takes these differences into consideration.


Results indicate that calendar spreads tend to bear more risk, in the form of higher kurtosis than outright futures, after adjusting the two positions for comparable values-at-risk. These relative risks should be taken into account by practitioners when trading spreads, and by exchanges when determining relative margin requirements.


The remainder of the article proceeds as follows. The next section provides background and motivation. The third section describes the data and methodology. This discussion is followed by our results and conclusions.



BACKGROUND AND MOTIVATION

During the 1980s and 1990s the strong and sustained growth in U.S. equity prices dampened enthusiasm for alternative investments. However, the recent reversal in equity markets has generated a renewed desire for diversification into other areas.


One approach that has been striking a nerve with institutional investors is to allocate funds to a commodity trading advisor (CTA) or pool operator (CPO) who, in turn, divides these funds among a population of professional futures traders that specialize in various commodity markets. These practitioners rely on various futures trading strategies to actively manage their exposures to different commodity prices. The objective is for these traders to generate respectable rates of return that are largely uncorrelated with traditional portfolio holdings, thereby lowering the overall portfolio risk without significantly altering the expected return (Brorsen and Lukac [1990], Edwards and Liew [1999], and Schneeweis et al. [1996]). For example, many professional futures traders follow a disciplined approach of diversifying across a broad array of asset markets driven by different sectors of the world economy that are relatively insensitive to one another over time.


As a complement or alternative to diversification, many professional futures traders attempt to enhance returns by operating with rigorous trading discipline, for example, by ensuring that losses are constrained on unprofitable trades. For many traders the source of this discipline is reliance on an objective trading rule designed to terminate positions on losing trades while allowing profitable trades to remain in effect. The concept, at least, is quite simple; if you win on half of your trades and lose on half, you still gain if the amount associated with winners exceeds that associated with losers.


Generally, these objective trading systems fall under one of two approaches: trend-following or mean-reverting. In trend-following schemes traders develop a systematic, objective rule for discerning the evolution of a trend, which they then use to signal the likelihood of a profitable opportunity. When the objective criteria implied by the rule are met the trade is initiated and, as long as the criteria are maintained, the trade stays in place. In mean-reverting approaches, the rule is designed to determine when prices have “gone too far,” and some adjustment is expected back to a “more normal value.”3


Futures Calendar Spreads

An alternative to trading futures contracts outright is a strategy that speculates on relative futures price movements, by simultaneously buying and selling related futures. One common construct that satisfies this objective is the futures calendar spread. A market-neutral calendar spread employs two offsetting futures contracts (i.e., one long and one short) having the same underlying instrument but different expirations.4 For example, a trader can sell a short-term futures such as the nearby (next-to-expire) contract, and simultaneously buy a longer-term futures such as the next out (second-to-expire) contract on the same underlying commodity. For consistency with prior work, we define the value of a long position in this calendar spread as (F2 – F1), where F1 is the price of the nearby contract and F2 is the price of the next out contract. This position will increase (decrease) in value if the spread between the two futures prices rises (falls).


In some cases, spread prices (F2 – F1) are highly correlated with outright futures prices (F1 or F2), and traders might therefore use calendar spreads as a less risky surrogate for outright futures positions. In other cases, however, calendar spread prices exhibit low correlations with outright futures prices. Under these circumstances traders might use calendar spreads as a complement to outright positions to diversify portfolio risk.


Comparison of Risk and Return

Comparison of the risk and return performance of outright futures versus calendar spread positions requires that we clarify what is meant by “risk.” One important aspect of risk is captured by the standard deviation of price changes. Other critical attributes of risk focus on the behavior of the tails of the distribution, and thereby reflect the best and worst possible outcomes associated with an investment. These other attributes are embodied in higher order moments of the distribution of price changes, and include skewness and kurtosis. A normal distribution displays unique behavior with regard to the tails (no skewness and kurtosis = 3). Departures from normality might include, for example, downward skewness and thicker tails (higher kurtosis), which would both mean a higher probability of great loss than is implied by the normal distribution.5


Value-at-risk (VaR) technology offers an alternative measure of risk exposure that focuses on the left tail of the distribution. The VaR of a position is simply a measure of the maximum amount one would expect to lose over a certain period with a given level of confidence, assuming the underlying distribution of price changes is normal (Hull [2002]; Jorion [1997]).


If we assume that daily price changes in both outright futures and calendar spread positions (ΔF1 and Δ(F2 – F1)) follow stable normal distributions, it is a straightforward exercise to determine the number of calendar spreads that offers a comparable VaR to a single outright position, ex ante. In this regard we define σΔF1 as the standard deviation of daily price changes in one outright position, and σΔ(F2 – F1) as the standard deviation of daily price changes in one spread. The ratio, (σΔF1/σΔ(F2 – F1)), then gives the appropriate multiple of spreads to outrights that yields equal VaR, ex ante. We call this multiple the ex ante VaR-adjusted spread position.


Complications

Under the normality assumption for both ΔF1 and Δ(F2 – F1), the VaR-adjusted spread and a single outright would have the same expected maximum loss on any given day, for any given degree of confidence. However, ex post, these two strategies might experience divergent results, due to: 1) inherent instability associated with maintaining the VaR-adjusted spread, or 2) departures from normality in the futures or spread price change series. Consider each complication in turn.


First, the relative volatilities of these two positions, (σΔF1/σΔ(F2 – F1)), may change over time. Since a single outright futures contract is the benchmark portfolio, the VaR-adjusted spread must be rebalanced periodically to maintain the same ex ante VaR as a single outright at any point in time. As the ratio of standard deviations changes, periodic rebalancing of the VaR-adjusted spread represents an attempt to continually “hit a moving target.” Moreover, the VaR-adjusted spread will vary over time because each successive computation of relative standard deviations is only a point estimate of the true ratio. This implies that periodic rebalancing represents a source of inherent instability in the VaR-adjusted spread relative to a single outright.6


Second, under the joint assumption that daily changes in outright futures prices (ΔF1) and spread prices (Δ(F2 – F1)) are both normally distributed, their respective standard deviations are sufficient to characterize VaR. However, these two assumptions may not be jointly compatible. For example, it is common to assume that daily futures price levels are lognormal so that futures returns are normal. In this case, however, the spread price level would be the difference in two lognormals, which unfortunately is no longer lognormal (Poitras [1998]). Furthermore, we are interested in daily changes in the levels of both the futures and spread price series, rather than daily returns.7 Under the lognormal assumption for outright futures price levels, the daily change in the outright price is also the difference in two lognormals, while the daily change in the spread price is the difference of the difference in two lognormals. In this case neither price change series is well behaved.


In light of the growing body of evidence suggesting that futures price levels are not lognormal (see, e.g., Cornew et al. [1984] and Hudson et al. [1987]), some researchers explicitly assume that spread price changes are normally distributed, and then investigate the validity of this assumption. For example, Kim and Leuthold [1997] and Poitras [1990] document that daily price changes in futures calendar spreads experience significant departures from normality, in the form of both skewness and kurtosis. They suggest these departures may be due to the relatively small price changes that normally occur over daily intervals, and they demonstrate that lengthening the interval to consider weekly price changes results in fewer departures from normality. Kim and Leuthold [1997] then go further to determine the best-fitting distributions of spread price changes. While the distribution of futures spread price changes is close to normal (the normal is typically either the best or second best empirical distribution describing spread price behavior out of the 25 distributions investigated), they find the logistic distribution often fits the data better than the normal. On the other hand, their results are sensitive to the underlying commodity, sample period, sample size, and spread length, as well as the differencing interval, making it difficult to generalize their results.


While prior evidence is mixed, the possibility that futures and spread price changes are non-normal complicates our application of VaR technology. The analytical problem has to do with the methodology for determining VaR when the underlying true distribution is non-Gaussian. There are several ways to deal with this problem. Some are computationally cheap and some are not. They are all non-parametric and thus run the risk of efficiency loss in estimation. It is therefore safe to say that these methods might not improve upon the (possibly misspecified) Gaussian approach.8


While these alternative procedures hold promise for future research, we adhere to the simple approach of most previous work on these issues by assuming that outright and spread price changes are normal. We then investigate the validity of this assumption. The ultimate focus in this study is whether the ex post assessment of risk associated with VaR-adjusted spreads differs in any systematic way from the ex post risk of a single outright futures position.


DATA AND METHODOLOGY

This study uses daily data from 1991-1997, for 10 of the most active futures contracts traded in U.S. markets: crude oil, corn, Deutschemarks, Eurodollars, gold, live cattle, natural gas, the S&P 500, soybeans, and U.S. government bonds. The analysis proceeds in three stages.


First, for each contract we assess the nature and stability of the means, standard deviations, and correlation between daily changes in the value of one outright contract and one calendar spread position:(μΔF1, σΔF1), (μΔ(F2 – F1), σΔ(F2 – F1)), and ρΔF1, Δ(F2 – F1). In this analysis we wish to avoid potentially aberrant behavior associated with illiquid futures markets, or with expiration-related trading strategies. To deal with the former concern, we limit our attention to the most liquid nearby outright futures contract, and the spread position between the nearby and next out contracts. To deal with the latter concern, we roll over to the next contract maturity/maturities on the first day of the month in which the nearby contract expires. Thus, for purposes of this analysis, one contract period extends from the first day of one expiration month until the first day of the next expiration month. For each commodity contract, the analysis is performed for all nearby contract periods that expire between 1991 and 1997.9 We then aggregate results across all contract periods that expire during every calendar year, and we summarize annual results for the years 1991-1997, as well as over the entire seven-year period. Finally, we assess the stability of each standard deviation, as well as the ratio of standard deviations (σΔF1, σΔ(F2 – F1), and σΔF1/σΔ(F2 – F1)) across consecutive time periods. This first-stage analysis therefore sheds light on the nature and stability of return, risk, and co-movement in price changes for a single outright position versus a single calendar spread.


Second, the ex ante VaR-adjusted spread position is constructed for all business days over each contract period. Beginning on the first day of expiration month, the standard deviation of daily price changes is estimated for one outright contract and for one spread position, respectively, using data over the prior 60 trading days. The ratio of these two standard deviations determines the number of calendar spreads to hold on that day for the VaR-adjusted spread position. The VaR-adjusted spread is then rebalanced (i.e., the ratio of standard deviations is recomputed) weekly over each contract period. The ex post daily performance of this ex ante VaR-adjusted spread strategy is then examined for all contract periods over the years 1991-1997. Various aspects of the risk and return for this VaR-adjusted spread are then compared to those of the single outright position.


Third, a simulation is conducted that applies a naïve trading rule for each strategy—first to a single outright position and then to the VaR-adjusted spread. These rules simply compel entering a long position in one outright contract when a 20-day moving average of past outright futures prices rises, and entering a short position when the moving average declines. Analogous rules employing a 20-day moving average are also prescribed in connection with trading calendar spreads.10 The simulations cover every business day over the years 1995-1997. This final analysis pursues an alternative avenue to reveal, once again, whether the proposed VaR-adjusted spread strategy could have been used to achieve a similar level of performance to one outright futures position, in terms of return, standard deviation, skewness, and kurtosis.11


EMPIRICAL RESULTS


Ex Post Distribution for One Outright versus One Spread Position


Exhibit 2 provides the means, standard deviations, and correlations between daily price changes in a single outright and a single spread position, for all 10 contracts investigated. Results are summarized for annual sub-samples from 1991 to 1997, as well as over the entire sample period.


First consider the correlations between outright and spread price levels and price changes, respectively, presented in columns 6 and 7 of Exhibit 2. For many contracts these correlations are small in magnitude, and they vary substantially across periods considered and contracts examined.12 These results indicate that the calendar spread does not generally serve well as a surrogate for trading outrights. Instead, the low and unstable correlations documented in Exhibit 2 suggest that including spreads in a portfolio could provide risk-reduction benefits associated with diversification.


Second, consider the mean and standard deviation of daily price changes in a single outright versus a single spread, listed in columns 1 through 4 of Exhibit 2. As expected, for all 10 contracts, the mean or standard deviation of a single outright futures contract typically represents a substantial multiple over the analogous mean or standard deviation associated with a single spread position.