By Simon Kerr
One of the manifest difficulties of a successful trader or investor with their own firm is to diversify investment strategies. The monied trader will be very good at one particular way of addressing markets and won’t necessarily be a master of the universe in another investment discipline. The edge may be different and the money management necessary to be profitable will not be the same as for the core investment strategy. A common commercial strategy is to add capacity in a closely related investment strategy where there is a natural understanding and affinity. So a long/short equity manager whose process is based on fundamentals will hire portfolio managers to invest in a related market but distinct in capitalisation or by sector specialisation.
Where the core investment process is systematic, that is where it is based on trading rules coded into computer programs, it is natural that the added investment strategies are based on quantitative approaches to markets. The largest category of systematic managers is CTAs. The dominant style of investment of CTAs is trend-following – even now, when CTAs manage as much money as any other alternative manager category, trend-following contributes over two thirds of the returns of CTAs. This has made it particularly difficult for CTAs to diversify across investment strategies, but they do try by adding high frequency trading, cluster trading, statistical arbitrage and systematic strategies for individual equities.
Robert Rotella runs a successful and well known CTA business. His firm Rotella Capital Management has been around since 1995, and has some storied antecedents. Founder Robert Rotella worked for Commodities Corporation – the same firm that at various points housed Louis Bacon, Bruce Kovner, Michael Marcus, and Ed Seykota. Rotella had refined his own (directional) approach to markets over many years, and first launched his own fund in January 1998. In the Noughties he was looking to develop new systematic strategies across markets, but which did not utilise trend- following techniques. The intention was to diversify the investment processes of the firm with a systematic theme, and through that to give the firm internal diversification. Rotella did not want his firm to be beholden totally to trend-following.
So it was natural that it wasn’t long after he joined the firm (in 2003) that Veeru Perianan was allowed to pursue a research interest of his – to trade markets systematically using a long-short pairs strategy. Such relative value investment strategies are unlikely to have return outcomes with attributes like those of directional strategies.
Veeru Perianan |
In order to create a return stream that has different characteristics from traded equity and bond markets Perianan has developed trading systems that take positions on the relative price movements of instruments that have deeply liquid markets. The aim is to have a return stream that is uncorrelated to mainstream asset class level indices, including many hedge fund strategies, but the opportunity is better than that because by trading ratios and scaling positions appropriately the level of risk assumption can be finely controlled.
A sample of the sort of instruments included in the pairs of the strategy is given below.
Table 1: Sample Instruments By Asset Class*
Currency |
Equity |
Interest Rate |
Volatility |
IMM Australian Dollar | SGX MSCI Taiwan | EUREX Euro-Bund | S&P 500 Vol Index (VIX) |
IMM British Pound | OMXS30 Index | Long Gilt | |
IMM Canadian Dollar | FTSE/JSE TOP 40 | CBOT 30-Year Bond | |
IMM Euro FX | FTSE 100 Index | SFE 1-Year Bond | |
IMM Japanese Yen | S&P/TSE 60 Index | EUREX Euro-Bobl | |
IMM Mexican Peso | E Mini S&P 500 Index | CBOT 2-Year Note | |
IMM Swiss Franc | Mini Russell 2000 Index | 3-Month Eurodollar |
*Example only. Not necessarily representative of current portfolio.
When in the research phase for this approach Perianan wanted to establish which pairs of markets had relationships in the real world, i.e. not just those with a short term statistical relationship, but those which were related in a medium-to-long term fundamental sense. So the pairs were constructed within asset classes – such as one equity market index to another. “I wanted to determine which market instruments could be trusted together,” he says. “The aim was to create synthetic time series, and identify which were going to be the most useful – the most reverting time series; or the ones with an oscillatory pattern.”
In order to establish which pairs had meaningful relationships Perianan quickly moved on from measures of correlation – whilst this statistical measure is common it is very limited. In particular it does not give levels of conviction on causality – so spurious “relationships” between time series of returns can be given. Also the time-frame over which Perianan was looking to trade (days to weeks) was not suitable for the use of a derived correlation matrix – it would be too unstable and subject to too much change to rely on. Instead he applied co-integration to his research phase and subsequent investment process (see box on next page).
Perianan explains: “I was looking to establish which pairs had a stable consistent relationship. In other words I was looking to answer the question “are these two markets travelling together in the same direction?” Co-integration gives you methods to substantiate or disprove the investment hypothesis. For example, you can tie in behaviourally the relationship between large cap and small cap equity performance. We know there are leads and lags in how they move together. What we do in an example like this is look for the co-integration alpha that gives the most mean reverting, or the best linear, relationship. Using these methods to look at the spread of the two data series allow us to see if there is a stationary relationship or an oscillatory relationship.”
Co-integration (from wilmott.com)
Unlike correlation, co-integration refers not to co-movements in returns, but to co-movements in raw asset prices (or exchange rates or yields). If spreads are mean reverting, asset prices are tied together in the long term by some common stochastic trend, and we say the asset prices are “co-integrated”. Since the seminal work of Engle and Granger (1987) co-integration has become the prevalent tool of time series econometrics. Co-integration has emerged as a powerful technique for investigating common trends in multivariate time series, and provides a sound methodology for modelling both long-run and short-run dynamics in a system.
Co-integration is a two step process: first any long run equilibrium relationships between (say) exchange rates are established, and then a dynamic correlation model of exchange rate returns is estimated. This error correction model (ECM), so called because short term deviations from equilibrium are corrected, reveals the Granger causalities that must be present in a co-integrated system. The fundamental aim of co-integration analysis is to detect any common stochastic trends in the price data, and to use these common trends for a dynamic analysis of correlation returns. Thus co-integration analysis is an extension of the simple correlation based analysis. Whereas correlation is based only on return data, co-integration analysis is based on the raw price, exchange rate or yield data as well as the return data. Traded market data (as commonly analysed by CTAs) is not generally stationary in nature.
The TEXO manager put a lot into identifying the securities with the strength of relationship he was looking for. The rigorous assessment took a year, but was considered crucial by Perianan: “the selection of the pairs can make or break a program that trades securities together,” he explains. Eventually he arrived at more than 90 distinct pairs which had a statistically significant relationship. The criteria for inclusion of the pairs in the working model include that the relationships are consistent through time and hold in the medium term as well as the long term. Each modelled relationship must work in all market conditions such that there can be a strong expectation that the aggregation of all the modelled relationships will also universally add value for a multi-quarter holding period by an investor.
The data series produced for each pair is a synthetic time series and the portfolio manager analyses this as if it is a market on its own. An example is given in graphic 1.
Graphic 1. Example of Constructed Return Series For Related Instruments*
*Example only. Not necessarily representative of current portfolio.
The constructed time series is analysed for trading opportunities just like single market instruments – looking for break-outs, trends and reversals. In the example shown below there was a potential reversal identified in the ratioed-spread. In the sample the synthetic data series took 5-6 trading sessions to move from one side of the normalised trading band to the other. This corresponds with the typical holding period for the convergence trades of Rotella TEXO Program of 6 days.
Graphic 2. Representation of Convergence Trade Methodology*
*Example only. Not necessarily representative of current portfolio.
Perianan categorizes the trade example given as a convergence trade. Spreads between co-integrated markets can experience mid-term drifts wherein a medium-term trend exhibits change. These mid-term drifts create trading opportunities as spread relationships move from one regime to another. The manager of TEXO considers this kind of trade a divergence trade. Divergence trades have a median holding period of 14 days.
In constructing the trading program, portfolio manager Perianan targets stable performance across multiple holding periods with low system-to-system correlations. The targets have been achieved to date: overall correlation between the Convergence and Divergence systems has been a de minimis +0.12, and importantly for limiting semi-variance, the downside correlations between the Convergence and Divergence systems has been -0.19. Risk allocation to each system is done periodically based on a proprietary metric that incorporates volatility and drawdown information – the risk control is inherent to the portfolio construction process rather than being an external constraint imposed at a later stage.
The split of the program’s exposures by system (convergence and divergence) varies through time. The variation of risk exposure by asset class is larger, though as it is not a control variable but an outcome of other target variables in portfolio construction and trade selection, that is not unexpected. So investors will sometimes see most of the risk budget applied to one asset class at a particular point (84% to equities in August 2012), and at other times a more even split amongst the asset classes (equities 34%, FX 29 %, and interest rates 37% in April 2012).
Drawdown is an objective function within the portfolio construction process – how deep it may be and over what duration. “Downside risk characteristics are critical in many ways,” maintains Perianan,” and they have to be part of the portfolio optimisation”. The forecast Sharpe ratio is taken account of in the optimisation, though the manager is at pains to point out that the whole area under the curve is assessed, and not just the maximum or typical loss.
The leverage of each system is dynamically adjusted to keep the individual system volatility constant. The target volatility at the level of the whole portfolio is approximately 8-12%, though, as a control variable, it may be varied in a managed account structure if there is client demand.
In the quantitative management of equity portfolios there are often limitations on position size and constraints on industry size and factor exposures, but money management at the position level is often absent. Instead whole portfolios are risk managed. The CTA world has come from a way of operating (discretionary trading) that had single instrument risk controls, and there is something of a legacy of that in modern CTA management. For Rotella TEXO there will be discretionary reductions in portfolio leverage on a scaled basis if the drawdown exceeds pre-determined levels of portfolio volatility, so there is an explicit top-down risk control, but also a bottom up one. The trading algorithms that generate positions also enter position stops of two sorts – the traditional stop loss, and some adaptive stops that help risk manage positions that have been open for a period.
When the disruptive markets of 2008 came about Veeru Perianan, like many quants, took a fresh look at leverage management within his process. The key questions were “when do you step in, and how much do you reduce risk assumption across the Program?” Being statistically literate, with an MS in Financial Markets/Financial Engineering, Veeru Perianan needed to encode this new consideration. He assigned probabilities to whether market conditions were suitable for the investment approach taken by the Rotella TEXO Program, using a Bayesian framework. This decision support tool is a Bayesian network – it facilitates the determination of the probability of an event given a set of evidence. The approach allows the organisation of knowledge into a coherent whole. In this network the nodes represent variables of interest and the links are the causal or informational relationships. Having defined the network, the structure learns the appropriate parameters and can make inferences on the back of new (additional) data.
Investors in modern CTAs are used to seeing a consistent level of risk assumption at the portfolio level, and usually a very consistent spread of risk at the next level of a CTA. So a CTA may always run with a forecast volatility of 12 or 15%, and may typically (or always in some cases) have a third of the risk budget allocated to financials, a third to softs and a third to metals. Rotella TEXO is not a trend-following CTA, and the risk assumption at the asset class level varies considerably (as previously mentioned), and it also varies considerably more than a CTA at the portfolio level. Leverage management is a key differentiator for this program and any peer group. To illustrate, the 95% CI 1-day VaR for TEXO can be 0.27, or it could equally be over 1%. To put this in CTA terms (since Rotella TEXO trades in futures) the margin-to-equity may be under 10% or it could be over 14%. Leverage management is considered by the manager to be one of the strengths of the program.
The outcomes from the efforts of the Rotella TEXO manager are shown in Tables 2 and 3.
Table 2: Composite Returns of Rotella TEXO Program*
Source: Rotella Capital Management
*Past performance is not necessarily indicative of future results. For more information regarding this performance record, please contact Rotella Capital Management.
The return of Rotella TEXO over 67 months of trading is 29.3%, or a compound annual return of 4.7% from 61.2 percent positive months. With an annualized standard deviation of return of 7.2%, that gives a Sharpe ratio (1% RFR) of 0.53 since inception.
The Program returns show a small positive correlation to the returns of the S&P500 index, the MSCI World Index, and the HFRI Fund Weighted Composite Index. The Rotella TEXO returns show a negative correlation to the returns from the Newedge CTA Index. Although that is measurable, it is visible to the untrained eye from the returns shown in Table 3 – in each of the first five calendar years the sign of returns of the Program is the opposite of the sign of returns of the CTA Index.
Table 3: Composite Returns of Rotella TEXO and Newedge CTA Index
Year |
Rotella TEXO |
Newedge CTA Index |
2012 |
-1.75% |
-2.68% |
2011 |
2.25% |
-4.51% |
2010 |
-0.28% |
9.26% |
2009 |
17.19% |
-4.30% |
2008 |
-5.61% |
13.07% |
2007 |
16.70% |
3.28% |
Sources: Rotella Capital Management and Newedge
In fulfilling the design brief so well Veeru Perianan has done something very rare in hedge fund management. Many hedge funds have failed to adequately match the required first moment of a return series for their investors over the last five years (that is absolute returns). To be able to deliver a desired higher moment (control of volatility) is a real achievement.
And which sort of investors will look to act on this fine record? The most natural fit is with investors with meaningful capital committed to CTAs and global macro managers. There are funds of funds which concentrate on allocating to those strategies, and Rotella TEXO is a great diversifier for those specialists. In addition those investors in alternatives that take (funds of hedge funds) portfolio construction seriously might consider allocating a portfolio role to Rotella TEXO, just as they allocate to tail risk specialists and short sellers for their contribution to (FOF) portfolio risk. Platform operators that like to offer a full range of hedge fund strategies should find few conflicts with existing fund offerings.
Institutional investors have non-investment factors to consider before allocating to specific hedge funds or investment programs. Rotella TEXO is an offering from an investment management organisation, Rotella Capital Management, with sixty employees, a long history, and a world class infrastructure. The backing of the TEXO portfolio manager by Rotella Capital Management is not just in data provision and technology – there is a full range of support functions available – and TEXO is one of many internally tested investment strategies that Robert Rotella has personally backed with his capital and made available to outside investors.
Rotella TEXO only invests in markets with big volume. This makes the investment strategy highly liquid now and very scalable beyond the current $72m in the program, including managed accounts. The source of alpha is relatively unusual amongst hedge funds, and the control of downside risk through the leverage management Bayesian network is a proprietary advantage.
Investors in hedge funds state via surveys that lack of correlation is becoming more important to them and is rising up the ranking of investment objectives. Through fulfilling an investment objective of its own (diversification from trend-following), Rotella Capital Management can now offer a program that can do the same for other investors.
The old era CTAs did not feel the need to justify trend-following intellectually. They tried to identify emerging trends in markets and ride them until they broke because that was thought by many to be the most profitable way to participate in markets. Efficient Market Hypothesis or not, trending was often observed so it was plausible to make profits because the phenomena happened. Modern CTAs, with much more statistical rigour, much more data, using emergent applied mathematics, newly published academic insights into markets, and top of the range hardware, commercially available software, and in-house written code have a penchant for describing their activity as the application of science to markets.
Just as David Harding says that Winton has compiled a body of work – “a body of serious statistical work,” is his exact quotation – so the progress of Rotella TEXO has been about compiling a rigorous vires in relation to the application of co-integration and a Bayesian network. The techniques and tools of co-integration have been around since the early 1980s, and were given extra prominence when the Nobel prize for Economics for 2003 was awarded to Robert Engle, who, with Clive Granger, wrote the definitive paper on co-integration. Veeru Perianan, in being interested in working on related instruments was drawn to use co-integration “because it is essentially a method to substantiate or disprove an (investment) hypothesis,” he says.
The Rotella TEXO manager puts it that “the whole of the investment process is elegant because of the integration of the signal generation, portfolio construction, risk management and drawdown control. To me, what we have created has been like putting the pieces of a puzzle together.” Now with a publicly visible 5-year track record we can all admire the visible pattern of the Program designed to diversify.
The program is available only to those individuals and entities that are “Qualified Eligible Persons” as defined under Commodity Futures Trading Commission (“CFTC”) Rule 4.7. Investment in the program involves significant risks. There can be no assurance that the program will be able to realize its objectives.