The current Epsilon strategy started life in 2012 and has maintained a consistent style through several corporate incarnations: Metori spun out of Lyxor Asset Management in 2016, and the strategy seamlessly continued after Lyxor merged with Amundi in 2022.

Lower drawdowns and faster recovery

Amundi Metori Epsilon Global Trends UCITS is one of the largest and longest running UCITS CTAs. Says Amundi Head of Hedge Fund Research, Bernadette Busquere, “I have known the Metori team since before 2008. We monitor the entire CTA UCITS and non-UCITS universe globally and find Metori approach CTA and trend in a slightly different way, which brings some diversification. They have a different return profile”.

With a circa 70% correlation to trend following indices, Metori is unmistakably a trend follower, but has generated better risk adjusted returns than trend benchmarks. Epsilon has shown better downside control: its worst drawdown was just above one standard deviation against nearly two standard deviations for the SG Trend Index. Strikingly, Epsilon has also recovered from drawdowns much faster: whereas the SG Trend Index had a six-year drawdown between 2014 and 2021, Epsilon’s underwater spell lasted just 17 months between late 2018 and early 2020. Epsilon has also broadly preserved capital during some recently more challenging periods for trend following, such as 2023, which saw many reversals including one of the largest setbacks in recent years in March 2023 around the mini-banking crisis.

A CTA boutique

As an organization Metori is also different from the behemoths of the quant world. Metori is an independent boutique, wholly owned by five partners including researchers, and solely focused on trend strategies across a variety of markets and mandates, with and without commodities, and some products also include Chinese futures. Says Busquere, “Metori have skin in the game. The key PMs and researchers are partners, which shows alignment of interest. They are independent and focused unlike some bigger firms”.

The firm’s name, Metori, signifies its systematic DNA. “Our company name was crafted with the help of a naming agency to reflect our firm’s philosophy. During brainstorming sessions with linguists, we delved into our core values. We mentioned our strong consideration for financial theory, and the importance of implementing processes with method. “Metori” was inspired by the words “method” and “theory”, which highlights our scientific and systematic approach to investing,” says Gaussel.

It is also different from other systematic and scientific managers in terms of philosophy and process: maintaining an asset class agnostic perspective on markets; achieving more powerful diversification through how it measures and defines correlation; being more selective about which markets to trade – and more opportunistic about when to expand and contract risk, and when to trade more or less often, all subject to careful consideration of transaction costs.

Asset class agnostic

Traditional old fashioned trend models, including some of the published rules of ETFs and alternative risk premia strategies, are binary and view each market independently, having either long or short exposure to every market traded, and applying correlation matrices later for position sizing. Metori would rather start with correlations to determine meaningful groups of markets.

Metori would only end up estimating trends at the single market level if a market was uncorrelated. In practice they view trends through prisms of two or more markets. Epsilon builds a spider’s web to map out some intuitively unsurprising correlations such as different equity indices in China and Hong Kong, and others that are perhaps less obvious such as Canadian equities and certain currencies in South America. The same models apply across asset classes.

There are no predetermined asset class weights, and models are deliberately blind to asset class labels. “Asset class definitions are arbitrary and ignore interactions between them,” says founding partner, Laurent Le Saint. An unsupervised statistical optimization jointly estimates trends and inputs to find the portfolio with the best risk adjusted returns.

There can be longs and shorts within the same asset class. At times in 2023, the portfolio was long of most US and European equity indices, but short of Hong Kong equities. It had also been short of several US bond maturities but long of Italian fixed income. It was naturally short of Yen against Euro and Dollars but had been long of GBP versus the greenback. In emerging market FX there have also been a mix of long and short exposures. In 2024 there have additionally been long and short positions within equities and bonds.

The priority is finding strong trends in clusters of markets, but there are also some controls on position and factor concentration and on other unintended wagers. The portfolio can end up with some accidental relative value or spread positions, but these are controlled and will not become dominant, and some sorts of non-trend positions are avoided.

Though Metori do not predefine any minimum level of asset class diversification, their performance is not overly dependent on a single asset class. They report well balanced asset class performance drivers over time. Asset class performance attribution has moved around with all four asset classes in the lead in some calendar years. Currencies contributed most in 2023, 2022 and 2015, equity indices in 2021 and 2017, short term rates in 2020 and 2018, and bonds in 2019, 2016 and 2014.

Correlation clusters and true diversification

The correlation approach is not just about finding trends, it is also core to risk management. Whereas correlation breakdowns can upend strategies that presume a fixed pattern of correlation, Metori automatically picks up changes in patterns of correlation within or between asset classes. Says Busquere, “We find their framework of correlation clusters is quite unique, very special, and really central to their model”.

For instance, before the Trump election in 2018, US equities, Mexican Peso and oil were trading as one trade. And before the UK Brexit vote in 2016, the FTSE 100 Index, the British Pound and UK Government bonds were also all one big trade.

It may seem counterintuitive to suggest that a smaller portfolio trading only a few dozen markets could have more independent bets than those trading hundreds, but that is exactly Metori’s assertion. “We have three times as many independent bets as a traditional risk budgeting approach. The model maximizes diversification compared to a risk budget model,” says Le Saint. Says Busquere: “They spread risk over many independent bets but are not over diversified. They focus on positions that make a difference”.

Trading fewer markets with lower cross-correlations allows not only larger individual position sizes, but also more gross exposure without increased volatility at times of high model conviction. “We participate in trends more selectively and that allows the models to be more aggressive. Typically, the fund might only have positions in 20 of the 45 markets traded. It has never been exposed to all 45 though it has sometimes come close,” says Gaussel.

Optimising time periods for analysis

Metori strikes a balance between responsiveness and statistical robustness to distinguish signals from noise. The optimum time periods are different for volatility forecasting, correlation, trend lookbacks and trend holding periods.

Shorter term volatility and correlation indicators react more quickly to changing regimes and reversals such as the Covid crash. Says Busquere, “They can react faster, and achieve a higher Sharpe ratio”.

Meanwhile trends are based on longer term lookbacks, which straddles typical industry definitions of medium term and long term. Notably, holding periods averaging 60 days for the UCITS are shorter than the lookback periods. Metori is medium to long term on lookbacks, but at the shorter end of medium term on holding periods.

Market selection

Metori uses two backward-looking indicators to gauge the opportunity set for trend. The Epsilon trend indicator is simply a one-year lookback Sharpe ratio for trend following, while the Epsilon correlation indicator measures diversification within the investment universe. 2023 was not a great environment with the short term reversals around the mini-banking crisis in March 2023 wrongfooting some managers.

Portfolio turnover and exposure dial up and down with the opportunity set. Round turns per million average 330 but this moves around: it was much higher in 2020, much lower in 2021 and still on the low side in 2022. In early 2024, turnover was very low because trend signals worked well.

Turnover is intentionally variable. “The reaction function needs to be non-linear. For long periods of stable trends there is no advantage in rapid rebalancing. But when a Black Swan event such as the Covid crash occurs, the model surges into action and can then cut risk much more aggressively,” says Gaussel.

The lowest levels of exposure were seen in the second quarter of 2020, when the model exited many markets, reduced exposures, and forecast volatility dropped to below 5%, around half the historical average of 11%. Margin to equity has recently fluctuated between high single and low double digits.

“A fixed volatility target forces you to increase the size of low conviction positions when you have weak trends. Our model has a cap on maximum volatility, but we will also trade below the volatility budget under conditions of highly correlated, rangebound markets, when we have little confidence in trends,” says Gaussel. The classic model of trend following as a convex straddle (dating back to Fung and Hsieh’s 1998 paper) is still a good blueprint for Metori. A scatter plot shows a V shape return profile with respect to the size of market moves in either direction.

Transaction cost analysis

Metori can be opportunistic in expanding and contracting the balance sheet, but the models are cognizant of rebalancing costs. This means that if models suggest only a marginal reason for a change, the expected alpha may not be enough to outweigh the certain cost of putting on a trade. The proprietary mean/variance optimizer is smart and sophisticated enough to avoid excessive churning for marginal benefits.

Product suite beyond the UCITS

“The success of our UCITS fund was the pivotal milestone leading to the creation of Metori in 2016. Since then, Metori has successfully launched its own funds, offering various formats and a diverse trade universe to institutional investors,” says Gaussel.

The core client base for UCITS includes venerable wealth managers and private banks. While the managed account business caters for larger institutional investors who insist on their own structure for larger accounts and want more intimate involvement in research.

Metori also runs AIFs, a US onshore vehicle and managed accounts, and onshore China strategies, all of which can include commodities. Metori has been active in China since 2014 as one of the earliest CTAs to trade local futures markets.

Outlook

The benign risk asset performance of recent years can easily make allocators complacent about risk, but many of the largest asset owners want to maintain a well-balanced portfolio. US pension funds such as California State Teachers Retirement System (CalSTRS) are allocating 10% to risk mitigating strategies, of which half is in trend following.