Quant hedge fund manager of the year: La Française Investment Solutions

Risk Awards 2019: Focus on correlations helps LFIS stand out in sector’s horrible year

Left to right: Arnaud Sarfati, Sofiene Haj-Taieb, Guillaume Garchery, Luc Dumontier

In a year for alternative risk premia funds that’s drawn comparisons with Viking end-of-the-world myths, La Française Investment Solutions managed to avoid the deluge, vindicating its approach of combining dozens of low-Sharpe – but low-correlation – strategies. 

It’s been “challenging” but also revealing, says Guillaume Garchery, the firm’s head of quantitative research and development and a senior portfolio manager. He attributes LFIS’s resilience to its emphasis on diversification and avoidance of “common implementation traps” in the building of its alt-premia strategies. 

The firm’s flagship alternative risk premia fund is up 3% for the year to November, compared with Societe Generale’s index of alt-premia funds, which is down more than 4%. In February, the fund lost a modest 0.5%. In October it was up 0.5%, despite a record bad month both for alt-premia funds and hedge funds more broadly.

To put that in context, 2018 has seen SG’s alt-premia index suffer six of its 10 worst months since inception in 2016.

The firm’s modus operandi of combining low-Sharpe strategies while minimising the correlations between them has proven its worth over the longer-term, too. LFIS’s alt-premia fund has a Sharpe ratio of 2.1 over the five years since foundation.

“If you stack strategies – even with a very low Sharpe, but fully uncorrelated – you can reach a Sharpe of two or higher,” Garchery says. “If you have a portfolio of strategies with an average Sharpe of 0.6 but they are 30% correlated, you can add as many as you want, but you won’t get to a Sharpe higher than one.”

LFIS invests in about 30 different strategies grouped in three broad families: ‘academic’ premia, such as value or momentum; ‘implied premia’, based on volatility and correlation that play on structural market flows; and arbitrage opportunities such as negative basis trades.

“There is almost zero correlation between families,” Garchery says. As a result, principal component analysis shows the risk in LFIS’s alternative risk premia strategy is explained by around 10 significant components. Risk premia funds that harvest premia from only academic factors such as value and momentum typically have about four or five.

Low-Sharpes favoured

The choice of low-Sharpe strategies is deliberate. LFIS thinks risks are easier to identify for such strategies, making for more informed allocation choices.

“If strategies are uncorrelated, then the allocation process depends on a more limited number of estimated parameters and is therefore more robust,” Garchery explains. Managing the risks in high-Sharpe strategies, conversely, is harder because – by design – they exhibit fewer drawdowns in backtests.

“A three-Sharpe strategy’s performance will go up in a straight line. It appears there is no risk,” Garchery says. But the risks are there, just harder to see. Selling credit default swaps or implied volatility before 2008 realised a Sharpe ratio of three but ended badly, he points out.

“You’re better off splitting such a strategy into smaller strategies that are de-correlated, so you can identify the risks and allocate accordingly.”

When backtesting ideas, therefore, the firm is looking for problems, Garchery says. “We want to see tough times, because by seeing those you understand the risks in the strategy you are holding. You are testing two things: performance, for sure, but you are backtesting risk as well.”

Of course, seeking out dozens of uncorrelated strategies across multiple markets – LFIS trades more than 30,000 different instruments – requires a huge amount of spadework.

The firm generates about 100 million prices a day for risk management purposes. It spent two years at inception putting in place swap documentation with 30 counterparties, as well as the pricing models and trading infrastructure to trade such a breadth of strategies. Garchery says the setup is like a having a mini in-house investment bank.

An example is the firm’s backtesting software – coded in-house and using LFIS’s proprietary database of 100 terabytes, including tick-by-tick price data. The software is designed specifically to replicate what banks have.

For implied premia strategies in particular, banks often pitch axes – trade ideas that would allow the dealer to recycle risk, says Luc Dumontier, head of factor investing and a senior portfolio manager. “You need the tools to independently verify the back testing from the bank is valid and fits your expectations,” he says.

The risk function, too, mirrors that of a bank, calculating value-at-risk and stress tests for each individual underlying. LFIS sets leverage constraints based on a more than 18-month look-back period, which assisted the firm in avoiding the worst of the February turmoil.

In February, LFIS was helped by its long-term view that low-Sharpe strategies will go through periods of low volatility when returns are harder to come by. As a result, the firm had resisted pressure – both internal and external – to add leverage at the start of 2018, says chief risk officer Laurent Minvielle. “We were tempted to do it, but we didn’t,” he says.

The firm will not scale strategies based on only a 12-month view of volatility “by construction”, Garchery says. “We know in our simulation that there will be times when volatility will shrink, and for good reason.”

LFIS was launched in 2013 by Sofiène Haj-Taïeb, previously deputy head of global markets at Societe Generale, and Arnaud Sarfati, previously co-head of cross-asset solutions at the bank. It runs a hedge fund and Ucits versions of its alternative risk premia strategy as well as multi-asset funds and credit funds. Total AUM was $12.4 billion at the end of June 2018.

So, what comes next? Looking forward, the firm’s aim for alternative risk premia is to add as many strategies as possible to the portfolio, so long as they don’t overlap.

That includes experimenting with machine learning and alternative data to identify new signals or augment existing strategies – partnering with French fintech company SESAMm, for example, which sells market sentiment analysis based on social media and news coverage.

LFIS sees lots of scope, though, in the work it has already prepared. The firm is actively screening about 100 implied premia trade opportunities, says Minvielle, but is only trading a third of them so far. It’s still waiting for a “good entry point” to trade the rest, he says. “We have tons of ideas.”