CTAs are a mature and transparent industry that gives a predictable and clear view of pay-offs,” states Altiq LLP co-founder and CEO, Dr Sam (Tobin) Gover. He argues that CTAs are in many ways easier to understand than other hedge fund strategies: they generally trade exchange-traded futures, which are linear, ‘delta-one’ instruments with no strange pay-offs and rarely any liquidity issues. Gover recognises that returns for most CTAs have been underwhelming since 2008, but points out that the longevity of the managed futures industry means that we have abundant data on them. On a multi-decade view, since the birth of the strategy in the 1970s, the post-crisis, QE-era years are only a small part of the full history – and look like something of an aberration.
Target Sharpe ratios
Over a 30-year timespan, long-term trend following strategies and funds have delivered Sharpe ratios of around 0.5 to 0.7. Gover accepts the widely held view that this is probably a realistic target for that strategy on a standalone basis. But trend following is only one of many approaches pursued by the Altiq Global Program, which Gover thinks could reasonably target a higher Sharpe, somewhere in excess of 1. He thinks that certain, less scalable models could generate better risk-adjusted returns, and is determined that Altiq should remain nimble enough to exploit these. “From the outset we resolved to maintain a small asset base compared with other CTAs – and we could not run over a billion,” Gover states. The relatively low capacity target, he thinks, allows Altiq to focus on “higher-quality forecasting models, with shorter average holding periods of below five days.” This swifter style of trading most obviously distinguishes Altiq from some bigger CTAs.
Both shorter-term and longer-term CTAs tend to have no correlation to conventional asset classes over time, but shorter-term CTAs are a more heterogeneous group – shorter-term CTAs are less correlated to one another than longer-term CTAs. Altiq’s correlation to short-term traders and to CTAs appears in the matrix shown in Fig.1, and is somewhat below the manager’s intended range of 0.3 to 0.4. Low correlations amongst the universe of short-term CTAs are entirely natural for co-founder and CIO, Dr Peter Ho, who explains “there is more data available for shorter term models, so there are more ways to slice and dice it for a greater degree of difference.” Altiq is different from some other shorter-term CTAs, elaborates Gover, in that “we offer a diversity of models, mixing trend with reversion, short-term trading, macro, and machine learning, and we also aim to add value through execution.” And in 2015 Altiq’s performance has outpaced one widely followed index of shorter-term CTAs by a massive margin: the SG Prime Services Short Term Traders Index ended 2015 down 5.07%, while Altiq made +19.78%. Yet the outperformance came from trading over the same sorts of time frames: “Our short-term models did well in 2015, though the short-term CTA index was down,” says Gover.
The Altiq team thinks that innovative models have helped them to outpace many peers and, indeed, innovation has been a constant feature of their careers. Their broad investment philosophy was learnt at IKOS, which was “one of the first European hedge funds to use quantitative methods,” says Gover – while Ho claims to have been “amongst the early adopters pursuing statistical arbitrage strategies in equity markets.” After completing their PhDs in 1994, Gover and Ho joined IKOS when it had fewer than 10 staff and stayed with the firm as it grew to several billions of assets; IKOS was once a regular fixture in The Hedge Fund Journal’s Europe 50 survey of the largest 50 hedge fund managers in Europe. Gover and Ho co-founded Altiq in 2009.
Dimensions of diversification
Diversification is the hallmark of the Altiq philosophy. “Markets are driven by human interactions, and it is very difficult to model that behaviour, so you have to build many different models,” explains Gover. Consequently, Altiq’s suite of models seems quite eclectic. They combine top-down and bottom-up inputs; both trend following and mean reversion signals; price data and non-price technical data, including volumes and open interest, as well as some fundamental, economic data.
Time frames for data analysis, and execution, span the shortest price intervals available, tick-by-tick, out to longer-term, multi-week periods. But this does not imply high-frequency trading-style holding periods measured in the milliseconds. Data at these frequencies forms part of Altiq’s analytical resources, but their trading windows typically range from a few hours to weeks, with the average being two to three days. Altiq can home in on much narrower time frames to fine tune execution, and have built their own fully automated execution systems.
Diversifying by time frame is hugely valuable because the correlations amongst the time frames are near zero. This complements diversification amongst the momentum, reversal and macro models. So three families of models, across three or four time frames, of hours, days and weeks, add up to a dozen or so buckets. Different types of models and markets can be emphasised or de-emphasised over various time periods. For instance, “Trend models tend to do better over longer time frames and mean reversion over shorter ones,” Ho explains. Naturally, some of the models will throw up offsetting signals, with the trend and reversion models the most obvious example. These are netted for execution purposes as the portfolio changes via a very continuous style of trading, with up to 1,000 orders and fills throughout the day, from open to close. But transactions costs act as a brake on the process, particularly for shorter-term models, and if transactions costs increase, the system effectively switches off many short-term signals in order to focus more on medium to long-term ones. Markets with lower trading costs, such as FX and indices, tend to trade faster with more short-term alpha.
The three families of models – trend following, reversion and economic analysis – should see roughly even allocations over time, though weightings will vary as they are a consequence of bottom-up signals. Each cluster of models might peak at 1.5 times and trough at 0.5 times its typical weighting. Attribution of performance by model varies over time. It is no surprise that the trend models started delivering in mid-2014 through the first quarter of 2015, as indeed they did for most CTAs. What sets Altiq apart from some CTAs is that when trend stopped working later on in 2015, Altiq’s other models – notably their short-term ones – kicked in and carried on doing well throughout the year. Altiq’s performance is shown in Fig.2.
The models draw inspiration from many sources. Altiq’s research team of five, including Gover and Ho, contains three other quantitative researchers – Dr Tim Thoroughgood, Vinod Hirani and Nicholas Davey, who is also the trading systems manager – making “five technical minds altogether,” sums up Ho. The team reads plenty of academic literature and industry research, in areas such as behavioural finance, general mathematical methods, engineering mathematical methods, and machine learning. This builds on earlier academic research, such as Gover’s PhD, from Imperial College, in Digital Signal Processing and Ho’s, from Oxford University, in Bayesian Estimation. “All of this research helps us to analyse lots of data efficiently, and we need to build a lot of models,” says Ho.
For instance, relative value models and trades can involve calendar spreads, such as curve steepeners or flatteners, or inter-market spreads within the same asset class, such as UK versus European stocks. But one relative value trade that Altiq avoids is carry trades. Given Altiq’s average holding period of days there is not enough time to clip a meaningful amount of carry and, perhaps more importantly, Gover observes that carry trades “have some nasty left tail risk.” He goes as far as arguing that carry has now become a form of beta that is very easy to access, rather than a source of alpha. Carry is an example of a strategy that looks great on back-tests up to about 2007, but has not done well since then and is anyway vulnerable to flights to quality during ‘risk-off’ phases. Gover also thinks that the carry trade is more useful for larger funds.
If carry trades are off limits, Altiq also aims to limit exposure in terms of position sizes, models and certain market risk factors, including equity markets, commodity prices, bond indices and the US dollar index. Decomposing positions into these types of factor exposures is designed to avoid an unintended concentration of factor bets, which can arise when multiple models point in the same direction. If a large factor exposure ensues, then the factor constraints can be seen as a type of overlay that could scale back some positions to avoid over-shooting the risk factor limits.
Despite the diversity of inputs and constraints, it is important “not to make the models too complicated,” stresses Gover. With all of these models, the Altiq view is that it is vital to avoid ‘overfitting’ models with too many parameters that may appear to explain historical behaviour, but might not do in future. So the Altiq managers view more ‘parsimonious’ models as being more robust.
Execution, technology and liquidity
The models are developed for trading liquid markets. Altiq’s main criterion for selecting its investment universe of 55 liquid futures markets across equity indices, bonds, currencies and commodity markets is that they must be liquid, developed markets and traded electronically. Though Altiq’s main job is to build forecasting models, for signal generation and portfolio optimisation, it also has models for execution. Bespoke technology is essential, simply because “you can never get off-the-shelf products that do what you want them to,” explains Ho, and he considers that building software trading systems is part of his job.
Technology is also germane to regulators’ focus on cybersecurity and business continuity issues. For instance, Altiq came under the umbrella of the Chicago-based NFA when the rules changed in 2013 to remove some exemptions. Altiq’s NFA registration, as a CTA and CPO, entails some ongoing requirements, including cybersecurity. Altiq is also prepared for the possibility that MiFID 2 or other rules may require a full audit trail of trades, time-stamped down to the millisecond. This is not a problem for Altiq’s systems, which are all automated anyway as part of a technology set-up that includes “remote servers, resilient and fault-tolerant data centres, and high-specification hardware,” explains Ho.
No amount of technology can guarantee uninterrupted trading under all circumstances, however. Altiq admits that liquidity has sometimes completely disappeared from certain markets, and the managers give three recent examples: the Treasury Flash Crash of October 2014, the Swiss Franc de-pegging of January 2015, and the Second Flash Crash of 24 August 2015. Gover freely admits that it was not possible to trade normally during these intra-day episodes. Separately, Altiq can exercise discretion to halt trading or downsize the portfolio.
Notwithstanding these occasional hiatuses in trading, the Altiq managers are confident about adding value through execution models, which have changed much faster than the portfolio construction models. Gover is pleased to say that Altiq has improved execution capability “after doing a lot of work.” He reckons that the advances here could be of the order of 1% per annum in performance terms. Altiq uses industry-standard FIX API Protocol for execution, and benchmarks ‘best execution’ against a range of metrics, including VWAP and the arrival price, which is the price prevailing when Altiq decided to make a trade.
Investment vehicles and regulation
Head of Marketing and Investor Relations, Giorgio Rosati, explains that in late 2015 most of Altiq’s assets were in managed accounts, as their predominantly US investor base prefers to access CTAs via these vehicles, for reasons including cross-margining and control of assets. In general Altiq would want to see a $10 million account size for managed accounts (at a 12% to 15% volatility target), but can accommodate smaller accounts so long as investors can cope with some degree of tracking error arising from trade size granularity issues. For markets such as Japanese government bonds (JGBs) the large contract size means a smaller account will either end up with overweight, or underweight JGB positions, relative to the model signals.
Though Altiq’s managed accounts are outside the scope of AIFMD, Altiq’s Cayman fund required the London-based firm to register as a full-scope AIFMD. This is partly due to the way in which AIFMD defines leverage as gross exposure, meaning Altiq’s gross fund assets exceed the €100 million threshold. Thus far Gover has found AIFMD to be the most time-consuming regulatory compliance obligation he has handled. Altiq’s compliance effort combines internal resources, from Gover and others, with external advice from compliance consultants Cordium, who assisted with the original FCA application in 2010.
Altiq has not yet sought to avail of the AIFMD passport, and is reliant mainly on reverse solicitation for European investors who wish to invest in the Cayman fund rather than a managed account. Capital introductions assistance is provided by Altiq’s prime broker, Societe Generale Prime Services (SGPS). Altiq became a client of SGPS via the latter’s acquisition of Jefferies in 2015, though Gover and Ho had known Duncan Crawford, Global Head of Hedge Fund Sales and Capital Introductions, Societe Generale Prime Services, for decades, and view the Soc Gen team as “very well respected brokers who have been around a long time.”
In addition to managed accounts and the Cayman fund, Altiq has looked at other structures, and Gover reckons a UCITS would probably make most sense as the firm grows. Altiq would be open-minded about launching a UCITS, or another fund wrapper, if an investor wanted a particular structure and could provide seed capital.
Since launching in 2011, Altiq has made changes to a number of models, but they should be seen as incremental additions rather than revolutionary changes. “We try to avoid tweaking models, as one rarely has enough data points to make changes very quickly for most models,” explains Gover. Altiq’s future research effort is directed at two main projects that have multi-month or multi-year time development horizons. The first is execution and short-term forecasting: “this is where you can add more value because you have more data and it is harder to build models,” says Gover. The second is enhancing the macroeconomic models. Macro variables already analysed include interest rates, and inter-asset class relationships such as those between currencies and commodities. Going forward “a big research project for us is looking at how we can use macroeconomic data on inflation, unemployment and growth,” reveals Gover. Though Altiq is already one of the top-performing CTAs in 2015, the firm continues to innovate through research.
Dr Sam Gover is a co-founder and partner of Altiq LLP. Prior to this, Dr Gover headed the portfolio management team of the Cyprus office of IKOS from 2007. Dr Gover joined IKOS in 1994 where he worked as a researcher in quantitative finance before becoming the fund manager of the IKOS futures and FX strategies in 1998. Dr Gover has extensive experience in automated trading systems, quantitative modelling and risk management, with a particular interest in futures, commodities and currencies. Dr Gover holds a degree in Electrical & Electronic Engineering and a PhD in Digital Signal Processing from Imperial College, London.
Dr Peter Ho is a co-founder and partner of Altiq LLP. Prior to this, upon completing his academic research, Dr Ho joined IKOS in 1994 where he worked as a researcher, then manager of the IKOS Equity Strategies Fund, and most recently performed the role of head of research. Together with Dr Gover, he was also responsible for IKOS’s flagship fund, the IKOS Equity Hedge Fund. As a pioneer in the electronic trading industry, Dr Ho was among the first hedge fund managers to apply automated trading strategies in equity and derivative markets. Dr Ho has extensive experience in the fields of quantitative forecasting, risk management, and automated trading. He holds a degree in Engineering and Computing Science and a PhD in Bayesian Estimation from Oxford University.