The indices have generated an average net return to risk ratio of 0.81 over the 15 years and the ACS programs have returned 0.85. These are high ratios for a systematic strategy over such a long period. The monthly return profile for ACS also shows a strong positive skewness coefficient, which is partly attributable to Aquantum’s emphasis on research and management of volatility and risk. The return profile is also uncorrelated to other systematic strategies, hedge fund strategies or conventional asset classes. This unique return pattern is not surprising given that Aquantum’s concept and way of trading seasonality is very different from other commodity strategies trading calendar spreads, which tend to focus on mean reversion or “roll yield” carry or both. Moreover, Aquantum is unusual in offering a pure play on seasonality, which is often one amongst many signals in other commodity and broader CTA programs. The increasing adoption of seasonality signals by other CTAs has not led to any changes in the Aquantum program.

Recurring patterns of seasonality

Aquantum has identified certain times of year when calendar spreads tend to move in a specific direction and the strategy will always trade in line with the expected move. Each market has its own seasonality characteristics. “The systems always know in what direction they plan to trade and would never trade in the opposite direction of the anticipated move. However, the size of the position varies with differing market dynamics and price action. It works on a systematic basis, and discretionary decisions based on current news are not used,” says Aquantum founder Thomas Morrow, who co-heads the trading team with Aquantum partner, Carina Polzer, who oversees daily trading. Two other team members are specifically dedicated to the commodity spread strategy. Morrow was formerly a senior scientist at Winton Capital Management after a banking career mostly in Australia, mostly at Deutsche Bank, where he developed and traded systematic strategies for the bank. “We know the logical drivers and past patterns suggest statistical significance around dates, even if we cannot pinpoint the exact date,” he adds.

Trades could turn out to be trend following or mean reverting with respect to the calendar spread, depending on the set-up of the curve at the time.

Constructing calendar spread trades

The structures involve longs and shorts in two or more different maturities in the same market. The spreads anticipate one part of a commodity curve outperforming, or underperforming, depending on the position of the spread. The shape of the curve when implementing the spread determines if a certain spread may be curve flattening or curve steepening. More complex calendar patterns involving two or more spreads in the same market may also be put on. “Two or more pairs could add up to something like a butterfly pattern, but each leg and pair works independently based on its own dynamics,” confirms Morrow. This expectation of over/under performance of a maturity is regardless of the directional price movement. There is no correlation between the underlying market price move and the performance of a spread.

Related markets such as soybeans, soybean oil and soybean meal are treated as three separate markets and the programs do not trade any intermarket spreads. Where seasonality coincides in two or more related markets, risk is systematically managed to avoid concentration of risk.

Fundamental reasons for moves

The reasons for the seasonal patterns are based on a variety of fundamental factors, including weather, global warming, natural catastrophes, inventories, storage costs, convenience yields and hedging behaviour. For instance, in the US hurricane season in August and September there is more likelihood of the front of the curve outperforming, regardless of whether the overall price is moving up or down. And the classic hedging premium applies: farmers pay a premium to hedge production around harvest times. “Around harvest time, there is better risk/reward being long shorter dated maturities than being short,” says Morrow.

Incidentally, climate change has seemed to essentially lead to parallel shifts in prices but has not changed seasonality patterns, according to Morrow. Northern hemisphere weather patterns are most relevant because liquid US-listed futures are traded, as well as brent crude and gas oil in Europe.

Aquantum want to see these sensible intuitive fundamental explanations behind the price moves. On the other hand, Aquantum have decided not to model the fundamental variables, and nor are they used to filter the seasonality factors. Aquantum’s research has found that there is too much noise, and that the fundamental data is already reflected in the price of each contract maturity. For instance, Aquantum has experimented with researching inventories, sentiment and other variables for signals but to date have not found added value from modelling them live. “Numbers are often released after the fact, too late to act on, and contain too much noise. It is all about expectations, which are most likely already in the price,” says Morrow.

Constantly re-testing theses

A nuance is that although fundamentals do not feed into trade signals, they are part of ongoing research designed to confirm if seasonality still exists. Indeed, Aquantum are constantly questioning their central theses, and exploring whether they might be invalidated by a range of factors over different timeframes. Short term shocks could include a pipeline closure or extreme weather and other idiosyncratic market events. Medium term events might cover legal, regulation and trading surprises; and long-term disruptive conditions could include global warming, trading limits or bans and mega trends.

“If spread moves are being changed by these factors, we could incorporate them, but to date we have not seen enough statistical proof to implement such factors. Every time there was a material impact, we reviewed the fundamental reasons and concluded that there was no reason to implement it. A one-off event cannot be included in the model,” explains Morrow. “The system will automatically adjust risk but would not adjust the models’ actual positions,” he adds.

In the research phase, supervised machine learning helps to gauge the stability of seasonal effects: “If the curves are no longer there, we might need to review models. Or it might just lead to a tweak such as trading one day earlier or taking advantage of statistically higher liquidity on Mondays and Tuesdays,” points out Morrow. Other techniques are also used to gauge calendar spread fatigue phenomena.

Covid and Russia’s invasion of Ukraine did not permanently change the relationships. “They did shift expectations and throw calendar spreads out of kilter, but they tend to come back quite soon. The programs did see a drawdown around March 2020, but all of them have since recovered it within varying timeframes. Such events do not change seasonality, but they inform our modelling of risk and volatility. We are always learning. We don’t assume we are always correct. The market price tells us if we are right or wrong, hence position sizes are adjusted when the price indicates we should,” says Morrow.

Investment universe: liquidity, seasonality and cost criteria

The same 25 markets have been included in the trading portfolio since inception. Feeder cattle was temporarily ceased but has resumed as liquidity returned. No new markets have shown enough liquidity to be included in the portfolio.

Liquidity in the second and third contracts is especially germane to calendar spread trading. “Less liquid contracts have wider bid/offer spreads and smaller size available on each side. They also reduce the signal to noise ratio. When a calendar spread moves, we want it to be due to seasonality not illiquidity. It is harder to discern patterns of seasonality in less liquid markets with high slippage,” explains Morrow.

If contracts are not liquid enough at any maturity, slippage costs can be too high. “We must always have liquidity. If a trading ban is temporary, we would take the market out. If a trading ban lasts a long time, we might have to permanently remove the market,” says Morrow.

Limited liquidity beyond the front month – as well as low levels of seasonality – explain why Aquantum has a very small risk allocation to cocoa, in contrast to directional traders.

Aquantum stopped trading metals for similar reasons: they did not show strong seasonality, and the LME was expensive to trade on (before and regardless of the controversy around cancelling nickel trades). Aquantum had traded the second Tuesday constant expiry LME metals but discovered that they were expensive in terms of slippage and other costs. “The small notional contract sizes are less volatile for leverage purposes, so big lot sizes would be needed, and they are expensive,” says Morrow.

However, improved liquidity has allowed for trading further out the curve in oil.

Geographic shifts in supply can render some markets less relevant over time. “One wheat spread has disappeared from the portfolio because the market share of US wheat, which was one the largest globally, has reduced as non-US markets have grown. Hence seasonality in US wheat has been reduced, as participants can now go elsewhere to obtain wheat,” says Morrow.

Selective trading and holding periods

On average the broader commodity calendar spread strategy is exposed to 12 of its 25 markets but can sometimes go up to 18 or 20 while ACS Energies trades 5 or 6.

The number of trades partly depends on contract periods. Aquantum trades the front four contracts, which are four months for energies but about a year for other markets. “Energy has 12 contracts per year, but other markets tend to have 4 or 5, tied to harvest times. Sugar has 4 contract months every year, and seasonality is 4 times a year as well due to the harvest cycle,” explains Morrow.

Average holding periods are 15 days but occasionally over 30 days and can run to two months for quarterly contracts. Therefore, Aquantum does not roll between contracts, as in a typical roll yield strategy. “Roll yields are considered and play a part in spreads, but we do not directly trade the roll yields. Roll yields are dependent on interest rate levels, and to trade an interest rate play investors should trade interest rates directly. We instead define carry as the interest rate cost of holding the position and research considers that. We are looking for outperformance irrespective of the interest rate environment,” says Morrow.

Model diversification and evolution of risk management

Though there is only one strategy conceptually, there are multiple models: “There have always been about six systems, and none have been added or deleted. There have always been dynamic take profit and stop loss targets,” points out Morrow.

The main refinements have been managing risk and volatility as the systems can now cut risk quicker than they did previously. Risk management springs into action once the trade is put on: “Risk management is more reactionary over the lifespan of the spread. The statistical significance of moves from the entry price helps to determine the risk management reaction,” explains Morrow.

A convex strategy

The approach anticipates a higher probability of profitability at specific times of year. It is not a reactionary program that awaits certain conditions before reacting. The hit rate varies by market but has averaged around 53%, which is similar to the S&P 500, the BCOM index and the SocGen CTA index. However, ACS has a higher win loss ratio than all of these and zero or negative correlation to them all. In addition, ACS is a more convex strategy, with less larger losing trades (< -2 and < -3 standard deviation daily losses) and more larger winners (>2 and > 3 standard deviation daily winners). It is thus very different from other strategies trading commodities or other asset classes.   

Leverage and margin

Average gross exposure has ranged between 3x and 18x. This sounds very high, but net exposure is near zero and correlations between the legs of a trade are very high.

Net margin to equity is between 5-15%. “On spread trades in the same market, there is typically a 50% margin discount on the second leg, without which margin would be 15%. The 8% is exchange margin and brokers apply some on top, taking total margin to 10-15%. The low margin gives opportunities to trade at differing levels of risk,” says Morrow. The strategy is well suited to notional funding through managed accounts. In a fund structure, it can be sitting on 75% or more cash, earning interest in USD.

Discretion could be exercised to correct a breach of risk rules around leverage, margin and volatility, but this has never happened.

All trading is now electronic, on CME and ICE, but some discretion can influence the timing of execution and choice of brokers. The utility function and optimization constraints for a spread strategy can be different from a directional one. While directional strategies taking a single position are often anxious to avoid being “front run” by index rolls, for a relative value strategy the priority is maximizing liquidity to put on two or more legs of a spread. “Commodity index and ETF roll dates can allow for greater liquidity around the 5th, 6th or 7th business day of the month,” says Morrow.

Vehicles

The strategy can be accessed through funds and managed accounts run by Aquantum, and structured products run by third parties, such as UBS, which has licensed Aquantum’s IP as an index used to underly structured products.

Most of the assets in the strategy are managed in structured products, which have run accumulated notional peaking at USD 3.5 billion, and currently around USD 1.5bn. Leverage on various sub-strategy variants and notes can range from 1x to 6x.

Aquantum view the strategy, which reports under SFDR 6, as neither positive nor negative on ESG, but ESG neutral. They do not see potential to report under SFDR 8.