There’s something in the water in Montreal, Canada. Montreal-based Innocap, a division of the National Bank of Canada that manages approximately $3b in hedge funds, announced earlier today that it has entered the field of hedge fund replication in partnership with BNP Paribas. In fact, they’ve quietly been managing a product since July and are now ready to go public with it. As you may remember, it was only last fall that fellow Montrealers over at Desjardins Group went public with a fund that uses a form of “distributional replication” very similar to that used by Professor Harry Kat in London (see related posting). We were the first to publish news of that offering and we are now pleased to bring you the first interview with the managers of National Bank’s new entrant – “Salto”.
Unlike Desjardins, Innocap is pursuing the more mainstream “factor replication” approach. But the firm believes its particular take on the process is unique. The fund is based on a mathematical idea called “advanced filtering” borrowed from, among other fields, missile interception. The fund is designed to track the MSCI Hedge Fund Composite Index. A companion fund, “Verso” is designed to have a -1.0 correlation with the same index.
The Co-CEO of Innocap, Martin Gagnon, and the firm’s Managing Director of R&D Pierre Laroche spoke with us earlier today about the product, its rationale and its Genesis.
Gentlemen, let me start by asking you what you believe is causing all the recent interest in hedge fund replication.
Martin Gagnon: Many providers have recently launched synthetic hedge fund products. I think it’s a natural evolution of financial markets to create an index-like, low cost alternative to the vehicles offered in the early years of the hedge fund industry. There is now a need for investability, liquidity, transparency and shortability. But the problem is that investable hedge fund indices have been generally disappointing so far. Research suggests that only 20% to 40% of hedge fund returns are actually “true” alpha. And this proportion is eroding very fast – especially if you account for the fact that the relative illiquidity of hedge funds deserves a premium. In many cases, the alpha generated in recent years is roughly equal to the fee structure needed to get it. So synthetic hedge funds returns represent a viable alternative.
In addition, these products are “regulator-friendly”. Ours, for example, will soon be UCIT III compliant in Europe
How does it work?
Pierre Laroche: Basically, we incorporated 20 years of academic research into a process designed to identify 15-20 “candidate assets” (factors). Then we apply our proprietary filter to come up with a list of about 10 assets. Every month, the “Salto” filter dynamically allocates to these 10 assets using long and short positions. While the actual assets can change from month to month, we found that about 10 was the sweet spot. Any less and model risk was increased. Any more, and operational costs became an issue.
What factors are currently in the portfolio?
MG: Today, for example, we’re long 7 factors and short 3. Allocations range from around -5% to around +10% each. I’ll give you a few quick examples. We’re long small cap growth ad emerging value and short US Dollar and large cap growth.
Why did you choose to track the MSCI Index?
PL: In our opinion, MSCI is clearly the best hedge fund index on the market. Without entering into an academic debate about its technical merits, it’s easy to see its strengths. It has better reporting standards and a more rigorous process than other indices. The MSCI Hedge Fund Composite Index has over 1,400 constituents and is less exposed to equity markets than other hedge fund indices. It also has a better risk-return profile that most other hedge fund indices and is managed by a highly respected name in the traditional institutional investment world. Basically, if you’re going to minimize the tracking error to an index, it might as well be a great index. We’re actually the first to license the MSCI Index for this purpose.
There are several offerings available that use a factor replication approach. So what makes this product unique?
MG: Factor based methodologies are the most prevalent form of replication and we believe they represent the best alternative to creating liquid, low cost, efficient synthetic hedge funds. These methodologies can be used to create both trackers and absolute return products. JP Morgan, State Street, Blue White, and Deutsche Bank have created alternative beta products that have absolute return objectives, but not really “replication” objectives. They use certain risk management rules to truncate payoffs or they tend to use expensive and complex options strategies.
PL: Our algorithm is more powerful than the linear regression models employed by many providers. It aims for a low index tracking error, rather than low volatility or high returns. We actually looked at the traditional linear regression approach, but as you know, it doesn’t capture abrupt changes in hedge fund exposures very well. So we turned to “filters” – an idea borrowed from diverse fields like machine learning, missile interception, and signal processing. One of the best known of these filtering processes is called a “Kalman Filter”. I won’t get into the details of our decision, but we eventually decided to use different, more advanced filters instead. Like linear regression, these filters are also somewhat intuitive, but are better able to capture abrupt changes in hedge funds’ exposures.
The result is that we have a lower index tracking error than all of the methodologies currently used in the market. The model handles the non-linearity of factors very well and it also accounts for the non-normal distribution of error terms. And contrary to linear regression, this approach is not affected by the multi-colinearity between factors.
Why did you choose the factor replication approach over the so-called “distributional replication” approach advocated by Desjardins and Professor Harry Kat?
MG: We think the distributional approach is certainly creative and unique and we’d be willing to execute such a program if a client asked us. But as you know, this approach doesn’t track month to month returns – only long-term pay-offs. While this may be academically sound, many of our clients told us they preferred to minimize month to month tracking errors instead. The distributional replication approach can also be difficult for investors to comprehend and the time needed to prove that it works can be beyond the tolerance of many investors. Finally, it can be complex to implement and needs to be calibrated on a daily basis.
What about the mechanical trading approach used by some providers where hedge fund returns are approximated by actual hedge-like trades?
PL: “Strategy mimicking” as we call it falls into realm of the absolute return strategies. We find that approach has liquidity constraints depending on the instruments used and biases since some strategies cannot be mimicked. So the results can vary greatly and will often depend on the risk management rules used. These funds are quite discretionary and tend to compete directly with funds of hedge funds. They also charge a performance fee. So they’re not really “passive” and therefore don’t tend to focus on a low tracking error.
What do you feel are some of the drawbacks of the factor replication approach?
PL: We definitely feel that factor-replication is the best approach. It’s transparent and liquid. But most importantly, it’s intuitive. It’s not perfect though since it can potentially experience a “lag” as the model is based on historical data. Therefore, it can be delicate to calibrate – making the choice of algorithm very important. We believe our model addresses these potential drawbacks.
Is the product currently live?
MG: Yes. The fund has actually been operating and available to investors since July. We have waited until now to announce the product so we could finalize our licensing agreement with MSCI. We believe the origins of Salto are quite unique. We’ve been investing in hedge funds for over 12 years. So we have a solid competency in identify alpha and (alternative) beta. The fund was initially developed for the proprietary investment needs of a bank and seeded with proprietary capital from National Bank of Canada. We think this makes our product unique among its competitors.
How has it performed?
MG: As far as our back testing is concerned, we don’t rely on a specific observation period and we take into account the turbulence level in financial markets. So we’re very comfortable with our pro-forma historical returns going all the way back to 1994. We’re very please with the low tracking error of our algorithm in these back tests. It has been approximately 3% per annum since 1994 and has fallen below 2% per annum recently. On average, the fund underperforms the index in absolute terms since the MSCI Hedge Fund Composite Index is a non-investable index. This means that it contains a healthy liquidity premium. But you also have to remember that this is an index of actual hedge funds and therefore does contain some alpha. No matter how good a replication algorithm gets, it obviously can’t replicate manager alphas. Still, the Salto model outperformed most, if not all, of the investable hedge fund indices on the market.
Gentlemen, thank you for your time today.