Tiber Capital’s Short-Term Trading CTA UCITS

20 years of adaptation

Hamlin Lovell
Originally published in the April | May 2019 issue

Tiber Capital was founded in 2011 but its origins date back 20 years to the launch of a research company in 1999. The firm traded only proprietary capital for over a decade and began accepting external capital in 2015. It has thus developed very differently from most CTAs and has also delivered a distinctive return profile. 

Short term traders show much lower correlations to one another than certain other types of CTAs, and a decent number of managers (some members of the SG Short Term Traders index and others, such as Tiber, that are not members) have delivered single digit annualised returns with portfolio diversification benefits in recent years. Tiber Capital has shown a slightly negative correlation to equities; near zero correlation to the Soc Gen CTA and Short-Term Traders indices, and to both indices’ individual constituents. The strategy has received The Hedge Fund Journal’s ‘UCITS Hedge’ award for best short-term trading CTA, based on risk-adjusted returns over 2017 and 2018.

Indeed, another distinguishing feature of Tiber’s return profile has historically been that the maximum drawdown – of 6% since 2014 – is much smaller than that of other CTAs, which have often seen 20-30% drawdowns over the past five years. Tiber’s low drawdown is partly a function of its low volatility, and when adjusted for this, the difference from other CTAs is smaller. The shallow drawdown is also related to Tiber’s positively skewed return pattern, which is more pronounced than that of many tactical trading managers. CTAs have shown positive skew over multi-decade periods, but over the past five years the skewness (using monthly data) of most names in the space has been negative. 

Tiber’s downside control, in turn, is partly due to strict stop loss rules that automatically exit positions – and can liquidate the whole portfolio – based on preset price and time metrics. Conversely, winning positions can run for longer periods. “The obsessive application of our risk rules makes it possible to have not only restricted downside volatility, but also adds very compelling positive skew and convexity,” says CIO, Mauro Taratufolo.

Tiber Capital has developed very differently from most CTAs and has also delivered a distinctive return profile.

Commonalities with other CTAs

Yet Tiber has some qualities in common with many CTAs and short-term traders. Of Tiber’s best four calendar years (2007, 2008, 2010, 2014) three have, in fact, coincided with strong years for traditional trend followers with 2010 the odd one out. “There is no doubt that low correlation among the underlying markets traded, increased volatility and good directional volatility (follow-through) are all favourable environments for us as well as for the entire CTA space,” explains Taratufolo.

Tiber is 100% systematic and 100% technical. “The models are purely price-based, using several types of market generated information (volumes, market internals/sentiment indicators, value concept) as additional confirmation,” he adds. Like most short-term traders, Tiber trades over timeframes ranging from intraday to multi-week. Also similar to many peers, the periodicity of data inputs ranges from minute to daily frequency. Additionally, the investment universe of around 30 futures has a strong overlap to that of other short-term traders. “It is selected based on liquidity, measured by volumes and open interest. For instance, we trade the five and ten-year Treasury – not the two year,” says Guglielmo Mazzola, who works in Trading and Research. The capacity target is also typical for a short-term CTA. “We are confident that the program is scalable to capacity of 1-2 billion, based on our research and simulations into market microstructure, execution and algorithms. At the same time, we realise that short term traders cannot run 10 billion, based on looking at our peers,” says co-founder and CEO, Sebastiano Zampa.


But Tiber has a larger weighting in equities and commodities than most CTAs: with 55% in world stock indices; 10% in each of grains, energies and bonds; and 5% in each of currencies, metals and softs. This adds up to 35% in commodities, and only 10% in bonds and currencies together.

Another defining quality is that asset class weights are fairly fixed, as are weightings to the the five families of models: Volatility Breakout; Momentum; Mean Reversion; Pattern Recognition and Tactical Trend. The weightings are based on a systematic methodology using a classic portfolio optimisation model, with a modified VaR objective function run monthly, subject to constraints setting minimums and maximums for asset classes, styles, and time frames. “Applying a more sophisticated and dynamic risk sizing approach does not really work because we are already very nimble and adaptable. Such techniques may work better for other strategie such as medium-term and long-term ones,” argues Taratufolo.

The five families of models in high level terms sound similar to other CTAs, but are tailored to asset classes. Some CTAs apply all models to all markets traded whereas Tiber only trades equity indices using all five families of models, while other markets only use a subset of the models. Commodities and bonds are traded only by Momentum and Tactical Trend (at Short Term and Intermediate time frames), while currencies are only traded with Momentum models. “The rationale for this is that we have strong market knowledge and academics including a detailed understanding of the intraday dynamics of market microstructure. This granularity of data lets us be sensitive to the specificities of instruments,” explains Mazzola. 

Academic and real world influences

The inspiration for Tiber’s approach comes partly from academic sources and ideas from successful discretionary traders. Work on selecting models, developing systems and constructing portfolios has also been used. The academic foundations should not be over-emphasised however. Taratufolo’s original influences were based on his discretionary trading, and he is under no illusion about the challenge for Tiber’s intraday models, pointing out that only 5% of intraday traders make money. Says COO, Luca de Lucia: “To be consistent and profitable in the long run, you want a good balance between in depth market knowledge, and a strict systematic approach, as a win-win factor in the long run. We often see a mismatch in the market between a highly regarded academic background, and proper knowledge of how the market functions. We spend a lot of time understanding how the market functions, and also apply strict rules.”

Incremental change and filtering

While some short-term traders revise and replace their models regularly, Tiber has changed its models to a lesser degree. Some 70% of its strategies have been there from the start of the strategy 15 years ago. “The core systems have been fine-tuned and improved – with some new systems introduced – but the change has been more incremental than revolutionary. The objective has always been to arrive at trading principles that persist over time. Investors do not like style drift. We have only one strategy and one product,” says Zampa. 

The research environment is independent and entrepreneurial, but also defined by clear boundaries of systematic, rigorous testing and validation, with no discretion. An example of how models are tweaked comes from the electronification of futures markets. “The advent of 24-hour Globex trading changes the psychology of peak session times, such as opening times, which were important in the old pit system,” continues Zampa. “This is not good for systematic trading approaches based on volatility breakouts seen at certain times of day. We have addressed this change in market dynamics by modifying our intraday systems.” Mean reversion techniques have also been improved with different exit techniques. New strategies are piloted for a few months using proprietary capital before going live in the fund.

The biggest change to the models came in 2014 with the introduction of filtering systems, which provide context for signals, have reduced the number of false signals, and cut portfolio turnover – from around 5,000 to 4,000 round turns per $1 million of assets. The filtering techniques have increased hit rates from a typical 40% to 50%, though this does fluctuate. Considerable effort was devoted to designing the filtering systems. “The challenge was to improve hit rates without losing the positively skewed and convex return profile,” says Taratufolo.

The filtering system is based on the same principles as the original models. “The battle we all fight is to separate signals from noise. No complex data mining, statistical learning, machine learning or artificial intelligence techniques are used. The filtering techniques aim to provide a road map for the day, based on volume and price movements referenced to previous values. We stick to simple systems, based on very few parameters, elegantly used, rather than optimising or changing parameters,” he adds. Thus, parsimonious models are the name of the game.

“Of course, there is still much complex mathematical and statistical coding and tools involved in testing and validating the data to find good and robust strategies,” clarifies Mazzola. “But,” he goes on to stress, “we end up with two or three new ideas each year, not two or three hundred.” 

No forecasting

Tiber’s philosophy is unusual in containing no forecasting elements, which seems conceptually striking. The usual assumption is that signals must be based on forecasts. Taratufolo explains, “The meaning of not forecasting is that we don’t use any mathematical/statistical methodology, which implies somehow any forecasting element. None of our trading models tries to forecast future price levels for the underlying markets. We prefer to describe our approach as reacting to what the market is doing and engaging the market in the direction where there is least resistance, applying adaptive techniques to manage and decide when and where to exit positions”. 

Though Tiber generally have some overnight positions from previous days, they begin each trading day with a fairly open mind. “We approach the market in a neutral way and do not start a trading session with a particular bias. Our main mantra is that we are good at reacting to whatever the market throws at us, rather than being good forecasters. The market is there to surprise you and will not show the same behavior as in the past,” he continues.

Attribution and market climates 

Rather than varying allocations to markets, models or trade timeframes, the objective is to generate “all weather” models that continue to deliver a positively-skewed return profile and protect portfolios. 

Attribution amongst the five models varies from year to year. For instance, 2015 and 2016 were better for mean reversion while 2017 and 2018 were better for momentum. Similarly, the contribution made by different markets moves around. The largest allocation, to equity indices, has actually made a positive contribution every year since 2014, but grains and softs have been profitable in some years yet not in others.

Tiber has produced some market regime analysis demonstrating the versatility of the strategy, which has, on average during 2018, generated positive returns during various economic climates of rising and falling: economic growth, inflation and interest rates; risk-on and risk off phases; and a category called “pension building” whereby equities and bond yields are up, with “pension busting” being the reverse. 

Over the past decade, Tiber have also been consistently profitable – throughout 2008 to 2018 – during what they term “VIX dislocations”: defined as the top 20 spikes up; all daily spikes over 10%; all increases; all decreases, all daily spikes below 10% and all daily spike above 20%. 

Tiber also views attribution through the lens of market regime analysis, which suggests that 2018 was highly unusual in having eight of the following nine regimes: bull volatile, normal and quiet; sideways volatile, normal and quiet, and bear volatile, normal and quiet. What was even more peculiar about 2018 is how violently and quickly markets migrated from one regime to another, often in a matter of days. “We can adapt and respond to changing market regimes and environments,” says Taratufolo. 

Taken together, February and October 2018 provide one reference point for comparing Tiber’s performance with traditional CTAs. According to Tiber, four trading days during these two months accounted for 10% of negative performance for CTAs, but Tiber was only down a total of 0.2% over the same days. Still, Tiber found February easier to navigate than October. “A higher VIX generally means expansion in volatility, which was profitably captured as in February 2018. If the VIX move is sustained, there tends to be more follow through into momentum models,” says Taratufolo. In February, Tiber’s models were nimble at cutting and reversing from long to short, and in locking in profits from shorter term models before regimes changed. The Tiber strategy was on average flat during days of dislocation, partly because intraday models caught the down moves. 

“October 2018, when Tiber lost 3.69% – our worst month since the filtering techniques were introduced in 2014 – proved more difficult because although volatility expanded, there was no trended-ness anywhere. The volatility was not directional in the right way, so we got whipsawed by false breakouts – for instance, the manager was long of equities on October 10th. Additionally, the market action was localised to equities, with no flight to quality seen in other asset classes. Intraday equity trading actually made money in October but the short and medium-term components did not,” he goes on. 

That said, no single day of returns in October was outside the range of expectations. It was a bad month due to the cumulative effect of many negative days. Tiber do not necessarily expect to profit in all market environments, but feel that October underscores the effectiveness of their risk management. 

Trading and execution

Tiber’s portfolio construction process may also be different from some CTAs, by steadily scaling in and out of positions, throughout the day, rather than making binary shifts. Trades are diversified by timeframe, and by entry and exit levels. 

There are also two dedicated execution analysts. Tiber’s execution methodology also dovetails with their anti-forecasting approach to signal generation: “whereas systems based on forecasting may need to execute bulk orders in a short space of time, we can be more patient because the architecture of our program allows the decision-making process to occur in a very fragmented fashion. For instance, if a trend day is unfolding, a diversity of entry and exit points will vary. We might hit peak margin to equity of 20% for only a few hours during the trading day – and be back down to a more typical 5% level by the end of the day. We believe that fragmenting signals reduces transaction costs,” says Mazzola.

“With an eye to AuM growth, we devoted a lot of resources and energy into developing our execution algorithms, aimed at improving execution with the increase in trade size. We upgraded our order management and execution infrastructure. We felt we needed to upgrade from the infrastructure we had while running the business like a ‘family office’. We implemented a new OMS and developed a new proprietary FIX execution engine to make sure the entire trading process was institutional grade,” says co-founder and CEO, Massimo Bochicchio.

But Tiber’s approach also has some refreshingly simple features. The research phase includes conservative slippage assumptions, factoring in three or four times the realised level of slippage. “This is a simplistic approach, but we do not have false expectations about signal quality or get nasty surprises from live trading. Our execution algorithms are in fact more based on market knowledge than on pure mathematics, and we have developed different methods for equities and softs for instance. We benchmark our own execution against various other third-party algorithms so we know the levels of acceptable and quantifiable slippage. In big picture terms we read that slippage has come down from 50% to 20% of gross returns, for many traders – and we are also seeing less slippage as a percentage of our returns,” says Taratufolo. 

“But in many markets, we do not need to use algorithms to stagger orders – we can sweep the whole order through,” he adds. 

“We are not co-located per se but our servers are in the US, closer to where most trades are executed. Millisecond or microsecond differences in latency are not really critical for our style of trading,” sums up Mazzola.

Institutional operational infrastructure 

Improvements to execution techniques were just one facet of an institutionalisation drive that paved the way for accepting external capital in 2015. Indeed, the operational side has changed more radically than the investment strategy. The firm spent 18 months building up the right infrastructure.

“In terms of the team, we made sure that we were structured enough to cover the key functions of an institutional grade investment company. We were a team of seven and had full control of all operations (back office, risk, compliance, research, investment, etc),” says Bochicchio.

“In terms of investment vehicle, when we first started marketing the Diversified Program (end of 2014), we were only accepting investment via Separately Managed Accounts (SMAs). The first ticket – in a managed account – was garnered at the Miami Context Summit in January 2015. One year later, an Italian pension fund that required a comingled structure provided the reason to launch a UCITS,” says Head of Sales and Marketing, Gian Rodolfo Errani.

Tiber’s UCITS sits on MontLake’s platform, which received The Hedge Fund Journal’s 2019 award for Leading European Management Company – AIF & UCITS Funds. The Montlake ML Capital platform, which has assets of $7 billion made good on its promise to set up a UCITS within eight weeks, and the product launched in May 2016. The UCITS, which has $85 million of assets as of March 2019, is run pari passu with managed account assets of EUR 190 million that include EUR 25 million of proprietary capital. Fees of 1.5% management and 15% performance are slightly below average hedge fund fees, according to the EY Alternative Fund Survey. Tiber are also present on the Allfunds Bank platform. 

Amongst other service providers, Societe Generale Prime Services (SGPS) stands out. Tiber has been with SGPS from the very start, in a relationship that began with SGPS forerunner, Newedge. “Their extensive knowledge of managed futures has been invaluable in terms of market knowledge and intelligence. They have been helpful with roadshows and capital introductions,” says Zampa.

Research evolution 

Future research is focused on two parallel avenues. The first is new models or markets, and the second is new filtering techniques to identify market regimes and context. There is also ongoing research into execution, mainly done by Anthony Leung, who joined from GMO LLC, and carries out research at Cambridge University. “There is a long bench of ideas in the pipeline and we expect to improve in all areas, but perhaps not massively and do not expect to be constantly adding new features,” says Taratufolo.

“The objective is not to deliver risk premia, carry trades, relative value or sell volatility in this strategy. We do not believe that this is the mandate of a CTA. We do not want to keep changing or drifting styles. We would rather stay committed to our approach,” says Zampa.

Tiber has started from similar foundations to many CTAs, but has designed and evolved its systems to generate a unique return profile.

Other strategies 

Tiber wants to maintain the purity and consistency of the short-term trading strategy, but might launch others as part of a “horizontal growth model”. So far, it has started one: a discretionary global macro cross-asset strategy run by Daniele Conte, who was hired from an Italian bank in late 2017. After some months of incubation, the strategy – called Tiber Evolution – launched in October 2018, in an AIF structure on the Montlake platform, and has assets of circa $30 million as of March 2019.

Evolution is run independently. “This fits in with our philosophy. We do not want to cannibalise or destroy the product as it is, and there is no overlap or interference with the CTA, but there are synergies in that the discretionary portfolio manager can benefit from risk management, data and number crunching capabilities,” says Bochicchio. 

Tiber could contemplate incubating other strategies, but is in no rush to do so, and expects to have three or four rather than ten.