European CTA pioneer, Man AHL, turned 30 this year. Over this period, managed futures – and Man AHL strategies – have generated better risk-adjusted returns than world equities, and with smaller drawdowns. In particular, managed futures have sometimes produced substantial positive returns, during crisis periods when equities crashed. The TMT bubble bursting between 2000 and 2003, the Russia/LTCM crisis in 1998, the credit crisis in 2008, the European sovereign debt crisis in 2011, are all good examples of how CTAs zigged when equities zagged.
There are sound reasons to think that trend-following alpha could be persistent and sustainable, over the long run. Man AHL’s Head of Client Portfolio Management, Graham Robertson, enumerates “behavioural finance teaches us that humans fall prey to biases, such as selling winners too early. Slow dissemination of, and under-reaction to, new information means it can take weeks or months to percolate into markets. And long-term macro cycles – both in terms of monetary policy and business cycles – move very slowly, again over multi-year periods.”
Man AHL’s 30-year life has coincided with a very persistent bull market in bonds, which most trend followers, including Man AHL strategies, have profited from, including when bonds rallied during market panics. Future market routs could of course see bonds and equities fall simultaneously. And Man Group has confidence in their potential to generate ‘crisis alpha’ from bonds as well as equities. In a September 2016 research paper entitled ‘Trend Following: Equity and Bond Crisis Alpha’, by Carl Hamill, Sandy Rattray and Otto van Hemert, Man Group demonstrates that momentum strategies would have performed strongly in the bond bear market between 1960 and the mid-1980s, when the US Ten Year yield rose from 5% to nearly 16%, and bonds returned less than the risk-free rate, as well as during the bond bull market since 1985. (As bond futures did not exist for much of the earlier period, Man AHL synthetically modelled proxies based on ‘cash returns, financed at the local short-term rate’). The hypothetical momentum strategy returns do vary according to the lookback period chosen to define trends. But it is widely known that medium-term trend followers use a one to four-month period (and in the paper, Man Group statistically infers that this time frame is the best fit for the BTOP50 index). Using this lookback period, annualised returns in excess of cash were in mid-single digits, and the strategy generated Sharpe ratio of 1.3, gross of fees and costs, over the whole period. The return profile also maintains the CTA hallmark of being a positively skewed one, akin to the payoff profile from buying a straddle, or both call options and put options. The associated ‘smile’ function shows that CTAs have done best in times of extremes – when equities or bonds were in their highest or lowest quintiles of performance. Notably, performance was best when equities or bonds were in their lowest quintile. As markets tend to fall faster than they rise, this is not surprising.
The 1960 start date chosen for the Man Group study is relevant partly as it marks the start of a 25-year bear market in bonds, but also because it ensures a reasonably broad dataset of reliable price data for many commodities. That matters because the returns from momentum strategies during bond bear markets have not only arisen from the obvious trade of being short of bonds. CTAs trade across multiple asset classes, which are inter-related. A bond rout can be manifested in powerful trends elsewhere. Most obviously if inflation is associated with the bond selloff, commodities will often benefit and sometimes show ‘parabolic’ price action. Currencies can also show sharp moves in risk off markets, and most of them only became tradable in the 1970s.
Man Group’s paper corroborates the findings of many other studies, such as Campbell and Company’s ‘Prospects for CTAs in a Rising Interest Rate Environment’, which is available on The Hedge Fund Journal’s website and covers a similar time period. Other studies go back for centuries, such as CFM’s April 2014 ‘Two Centuries of Trend Following’ research paper, which stretches back 200 years, and finds that all 10-year periods since 1800 have shown positive returns for trend following. Another study by Katy Kaminski of Campbell and Company, and Alex Greyserman of ISAM, suggests that trend following would have also worked well over 800 years.
If crisis alpha is expected to be the one constant, CTAs continue to evolve in several directions. Though CTAs on average, as measured by the BTOP index, have broadly gone sideways since 2011, Man Group points out that some CTA strategies have also performed well in the recent post-crisis period. Speaking at the SYZ Hedge Fund Day held in Zurich on October 31 2017, Man AHL’s Graham Robertson presented on ‘the evolution of trend following’. Today, Man AHL uses proprietary algorithms and momentum models to trade around 600 markets in futures, foreign exchange, OTC markets and cash equity markets. In the early days, there were fewer models, and fewer markets. Man AHL categorises the evolution of trend following CTAs into three thematic phases.
CTA 1.0: Moving Average Crossovers (MACs) and breakouts
Trends and momentum can be defined and measured in many ways. For Man AHL, CTA #1.0, starting in the 1980s, involved MACs and breakouts as the main signals. A typical MAC system used a ‘slow’ moving average, such as 200 days, and a ‘fast’ one, such as 50 days. Subtracting the slow from the fast, indicates either a simple direction, long or short, or a continuous signal of varying strength that translates into a position of varying size. Continuous MAC signals virtually always have some position, whereas breakouts are far more binary and opportunistic. Breakout signals trigger a position when markets move outside a channel defined around historical prices. Breakouts can get in and out very quickly, because the signal decays as the information dies away.
MACs and breakouts are different, and complementary. The combination of the two often results in better risk-adjusted returns than either alone. Some CTAs (such as FORT LP) have honed and refined their trend models, while others have added non-trend models. Man also runs non-trend strategies, mainly in different vehicles. But for trend following, Man AHL is happy to stick with MACs and breakouts and its researchers devote more of their time to other avenues of research. “Quants are good at prioritising their research time, and we think that searching for new markets to trade – rather than changing our trend models – is the best use of time,” says Robertson.
CTA 2.0: new markets
Historically, many CTAs maintained roughly equal risk in fixed income, commodities, currencies and stocks, focusing on a staple of up to 150 markets that are easy to trade via futures. Many CTAs still do trade this investment universe, and some of them have performed well post-crisis. However, some of the traditional markets have shown limited trends since 2008. For instance, interest rates in the G7 markets have been moving in lockstep with the US for years, whereas those in other markets have thrown up better opportunities for trading some types of trend-based models. For example, since 2009 Brazil has had three full interest rate cycles, resulting in five tradable trends: two uptrends, and three downtrends. In commodities, too, certain non-traditional markets happen to have recently displayed more trended behaviour. “European electricity has demonstrated persistent trends, while crude oil has often been somewhat jumpy and mean reverting, except in 2014,” says Robertson.
Other examples of non-traditional markets in fixed income include interest rate swaps, US agency mortgages, and options on bonds; in credit they include investment grade and high yield corporate debt; in currencies, they may be more emerging market forwards and options; in commodities, new markets include European electricity and coal. In equities, new markets have included sectors traded via cash equities, or ETFs, and baskets of stock. Sectors can be defined with standard GIC metrics. All of this means that Man AHL is effectively following trends in individual markets, clusters of markets, spreads between markets and risk factors.
Since 2005, Man Group has been adding these non-traditional markets and the evolution strategy trading them has returned 374% between 1st September 2005 and 30th September 2017, according to Man Group’s internet site. That compares with just 28% for the BTOP CTA index. Man Group declined to comment on performance. The outperformance has come from two sources. “Firstly, the standalone Sharpe ratios for trend-following in the non-traditional markets, have been higher than in traditional futures markets, over this period,” explains Robertson. In addition, correlations have been far lower in the non-traditional markets (and the traditional markets have seen a much sharper rise in pairwise correlations post-crisis, than have the non-traditional markets).
A low standalone Sharpe ratio for an individual market, model or strategy, is not in fact an insurmountable obstacle to generating strong risk-adjusted returns. So long as correlations between markets, models or strategies are low, it is possible to build a basket of them that has a relatively high Sharpe ratio. Some of the most consistent and long-lived systematic and quantitative investment strategies (including some trend-following strategies) are composed of components with Sharpe ratios as low as 0.2, which, combined with very low correlations, can generate an overall Sharpe approaching one. The new markets may be more lowly correlated where they are less susceptible to macro factors. For instance, hydro-generated Norwegian electricity has manifestly idiosyncratic drivers, such as levels of rainfall in Scandinavia.
There is of course no certainty that Evolution will outperform trend following in traditional futures markets in any or all market environments. For instance, this year trends in stock indices have been stronger than sector trends in aggregate, and conventional trend following in equities has outperformed Evolution’s sector trends. Conversely, when emerging market bonds, such as Brazilian paper, have been in a persistent trend – and developed market units, like ten-year Treasuries, were range-bound, Evolution has been in the lead.
Man Group is continuing its search for new sources of diversification through new markets, including harder to trade ones. “The richest vein of research in trend following is new markets,” reiterates Robertson. Many systematic managers are not thought to be trading these new markets. (The only other managers who have mentioned the benefits of alternative markets when we interviewed them over the past few years were Aspect Capital, Transtrend, Systematica and GAM Systematic (formerly Cantab)). Many of the markets mentioned above may require counterparty relationships, agreements and human traders. While Man AHL Evolution has stayed liquid, the Man AHL Evolution Frontier strategy – launched in May 2014 – has reached out along the liquidity spectrum and consequently offers quarterly rather than monthly dealing. Overall, Man AHL is adding 25-50 new markets per year. Robertson is not going to disclose the next markets to be added, but he will admit that Man Group has started to research bitcoin. “It entails challenges not seen elsewhere, including physical storage. This should be away from strong magnets!” jests geophysicist Robertson.
Labour-intensive trade execution in alternative markets is only one reason why Man AHL’s internal headcount continues to grow, with 150 staff, of which 100 are researchers. Man AHL also draws on the Oxford-Man Institute. Staff at the OMI are scientists who come from all over the world and can remain active in academic life. Sometimes, discoveries made in disciplines other than finance can be applied to finance. “Since 2016, the OMI has been shifting its focus from econometrics to machine learning,” says Robertson.
CTA 3.0: machine learning
The third episode in Man Group’s CTA journey has been machine learning. Though the concept of machine learning dates back to the work done in the 1950s, the discipline has become a source of more actionable signals fifty years later, due to three complementary developments. “Firstly, computational power has continued to grow as Moore’s law has proved its predictive power. Secondly, memory costs have fallen even faster than power has grown, with the cost per GB down from $300,000 to ten cents over the past thirty years. For instance, our data library now contains 1.5 trillion price ticks and is growing, sometimes as 120,000 price ticks per second. Thirdly, the methodology of machine learning has been refined as advances in statistics, computer science, engineering and mathematics have been combined to develop new approaches,” explains Robertson.
Machine learning can now sometimes generate signals pointing in the opposite direction of traditional trend following models. For instance, if a market has a short-term pullback, after a long-term uptrend, a traditional model might result in no position – as the short-term signal counterbalances the long-term one. But a machine learning approach could result in a long position being maintained. “The beauty of machine learning is that it is free form. It can be applied to all styles of investing, to mean reversion as well as trend following. Machine learning can also be applied to execution techniques,” says Robertson.
But the aim to profit in up and down markets has remained constant. The systems have also remained 100% systematic, with the exception of some discretion employed for manual trade execution in some markets.