“The rationale for a pure play trend follower was partly acknowledging some research showing that the trend factor could explain a high proportion of aggregate CTA performance,” says Philip Seager, Head of Absolute Return and portfolio manager of the ISTEC program at CFM.

Trend following differentiators and dispersion

It certainly seems simple to design a trend following model that looks strong historically, but the practical implementation of trend following can follow various approaches and there has been wide performance dispersion between strategies with similar names that include “trend following”. Some trend strategies have produced disappointing performance, particularly during equity market corrections including during the 2020 Covid crisis.

CFM’s trend models are one differentiator. “How trend signals are defined and measured can fall prey to overfitting and over-parameterisation, which may result in impressive back tests that do not perform out of sample. We like to minimize the number of parameters. We have gone through a lot of trial and error investigating many ways of exploiting trend following,” says Seager.

Portfolio construction is another variable. CFM’s trend strategies were also the first to use its proprietary “agnostic risk parity” techniques for more robust diversification, considering a range of factor risk exposures in addition to markets and asset classes.

The challenge of execution is a third consideration that should not be underestimated either since some simulations do not make enough allowance for transaction costs. “CFM’s trend strategies also make use of our specialist execution team, who are completely dedicated to looking at when and how to trade, and how to measure, model and minimize costs. All of this is subject to pitfalls in back testing,” says Seager.

These three research streams feed into CFM’s long term trend model, which is one building block for both the ISTrends strategy, which also has a cross-asset trend model, and ISTEC, which also has two other models: short term trend and a long volatility sleeve. Taken together, the blend of the three models in the ISTEC fund is subject to an overarching cap of zero on forecast equity market correlation.

The genesis of ISTEC, launched in 2019, lay in catering for different investors’ objectives for trend following strategies and as such it is perhaps surprising that the strategy did not start as one of CFM’s customized offerings.

Different flavours of trend following

Some investors just seek broad trend exposure, others prioritize broad diversification through a full cycle, while further groups of allocators specifically want convexity and protection for long portfolios of conventional assets during pullbacks or bear markets. Some investors became concerned that many diversified trend following CTAs (at least in the medium to long term category) had developed a structural bias to maintaining average net long equity exposure, and therefore did not offer as much convexity as investors might have desired. CFM is quite transparent about the net long equity bias observed in its own CFM ISTrends strategy: though the models’ net exposure would have been both long and short over the decades, it has averaged net long over the past ten years or so. 

Traditional trend followers, which trade all markets in both directions, do have an average equity correlation near zero over multiple decades. They might perform well in an extended crisis bear market such as 2000-2002 or 2008 but could be wrong footed by shorter lived equity pullbacks which can be called “corrections”. CFM has confidence in trend following over a 200-year simulation, but some clients want more predictable and reliable diversification over shorter periods. “Trend following has a positive drift due to behavioural biases, but it needs a more drawn-out timescale and did not work well in February 2018, when some trend following CTAs were long of equities ahead of a sharp correction,” says Seager. 

The path, as well as the speed, of a correction can influence whether trend followers profitably trade it. For instance, a “choppy” pattern of market prices can result in them changing direction and getting “whipsawed”. 

Three planks of protection 

ISTEC combines three strategies: long term trend following (7-11 months) using a mix of moving average signals; short term trends of around 2 months focused on the most liquid bonds and equity indices, and a long futures position in the VIX index of implied volatility. 

ISTEC employs three ways to increase the probability of providing protection. The overarching cap on forecast equity correlation should rule out a long equity market correlation, absent unforeseen changes in correlation patterns. And upon an equity market setback, both the shorter-term trend models, and long volatility exposure, should kick in more quickly than a traditional medium to long term trend follower. 

“The short-term trend strategy is also focused on trading a universe of contracts that provide more convexity and have lower trading costs in terms of slippage and market impact,” adds Seager.

VIX futures are seen as a purer play on long volatility than trend following itself. “While academic studies show that trend following mimics a long straddle position, in terms of producing option-like payoffs, owning actual volatility instruments can produce a faster payoff under a wider variety of equity pullback scenarios,” explains Seager. 

The choice of volatility instruments is an important part of implementation. “We trade the VIX future rather than options because it does not decay as much as individual options would,” says Seager. The front three or four months of the VIX are owned to optimize liquidity, not to trade calendar spreads on the short end curve shape (since there is anyway no short VIX exposure to create a spread). 

The aim is to provide potential protection over a range of timescales. Tactically short equity exposure combined with structurally long equity volatility are expected to be the main drivers of returns in equity setbacks. 

Covid acid test

Around the Covid crisis, the short-term model and the volatility model both made sizeable contributions to the overall program, which made 8.3% in March 2020, and 13.2% in 2020. All five asset classes: bonds, commodities, currencies, equity indices and short rates, were also positive in 2020, which is a high hit rate for a trend follower.

The equity cap: decomposing correlation and exposure 

The equity cap is defined as a zero cap on forecast equity correlation across the whole ISTEC fund. This need not always imply zero net exposure to equities, since other exposures inversely correlated to equities could offset any net long equity exposure.

The long-term trend model can go net long of equities, to the extent that other exposures reduce its forecast equity correlation to zero.

The volatility strategy is also constrained to have a less than zero ex-ante equity beta. It uses long equities to hedge long VIX exposure and mitigate the carry costs of volatility, and the ratio between the two is calibrated to a zero-equity correlation. The volatility sleeve always maintains long exposure to the VIX, and somewhat varies its hedge ratio with long S&P 500 futures, to maintain an overall beta around zero. Absent the long equities exposure, the VIX would be negatively correlated to equities around 90% of the time.

The short-term trend model is capped at zero net equity exposure. 

Correlation forecasting

Robust forecasts of correlation are clearly important for the strategy. CFM regularly monitors realized versus forecast equity correlation and has recorded a negative average for the program. “We model each instrument’s correlation to equities and clean up the data to reduce the noise, before constraining the portfolio to a maximum equity correlation of zero. Of course, these patterns of correlation are not always stable over time, and may themselves invert,” says Seager. For instance, the inverse correlation between bonds and equities over the past 20 years looks like an aberration in the context of a longer sample period and the two have sometimes moved in tandem during 2021.

Shaving the Sharpe to perk up the tail profile

The risk weightings to the three sub-strategies are currently fixed at 57% long term trend, 38% short term trend and 5% volatility. Though volatility is the smallest sleeve it punches above its weight and can make a big contribution as it did in March 2020. 

The weightings and equity cap are designed to strike a balance between enhanced tail risk protection and a reasonable long term Sharpe ratio. The strategy provides more tail risk protection, though this comes at the expense of a more modest Sharpe ratio in normal times. The back test shows a long run Sharpe of 0.4 since 1999, though it has been more than double that since the program’s inception, which included the Covid crash. Focusing on equity drawdowns, the back test shows consistently positive performance for ISTEC, which would have outperformed both ISTrends and the SG CTA Index.