This month, FactSet hosted a series of risk events throughout Asia and Australia. Representatives from Axioma, R-Squared, MSCI, SunGard APT, and Northfield joined us to debate current topics in risk management in six cities. We presented ideas, answered questions, and brought some new information to over 300 individuals representing a huge number of the major superfunds, pension managers, asset managers, and investment houses of the region.
In addition to presentations from all of the risk providers, our own Dr. Steve Greiner spoke about Currency Exposure Hedging and Stress Testing Potential Sovereign Default. The audience was then invited to ask questions to explore the different opinions and approaches of the providers. This lead to some entertaining clashes over the alternative approaches used and the reasoning behind them. I’ve summarized some of the questions here:
Q: How important is it for the risk model used to contain the factors used in the portfolio construction?
The panelists agreed here that the risk model should indeed include the factors that are being over/under-weighted, as without them it will be difficult to understand the inherent risks that these factors bring. This in itself brings an issue, as virtually all of the commercially available risk models have pre-specified factors, so any custom factors would be represented through a combination of those, an issue that had Laurence Wormald of SungardAPT and Jason MacQueen of R-Squared suggesting their own solutions. Nick Wade of Northfield highlighted a secondary function of a risk model in highlighting factors generating risks that perhaps hadn’t been understood or accounted for, not just those that had been built upon.
Q: Should risk models be daily in nature rather than monthly/weekly?
The panelists stressed the different areas that a comparatively simple sounding question does actually cover: use of daily data in models (against annual/quarterly balance sheet items, etc.); short term horizon models (designed for forecast volatilities over days rather than months); daily updates of monthly-built models. Ultimately, Neil Gilfedder of MSCIBarra underlined his belief that full daily updates are more responsive to changing market environments and this was the driving force for the recent changes in their approach. This methodology is already used by Axioma, and as pointed out by Olivier D’Assier, permits rebalancing at the behest of the manager, not dependent upon the update of the model vendor. It was nice to have some areas of agreement!
Q: How relevant are the risk statistics generated by the models at the constituent level?
Universally, all of the speakers agreed that each of the risk models on FactSet are designed to be used at the portfolio level and potentially provide some level of granularity at the sector level, but reviewing risk statistics at the security level can be dubious to say the least. The statistical techniques and estimation errors involved in the construction benefit from large universes and therefore to consider any individual security as being perfectly described can cause confusion and may lead to false conclusions. Large, well-diversified portfolios, benchmarks, and the like can be described accurately through factor models and therefore these should be the prime targets for their use.
The panel questions expanded to include potential data biases (“Is a Wednesday to Wednesday weekly delta better?”), use of forward looking data (e.g., options volatility surface), as well as the challenges of full data estimation (dummy variables for dummies, anyone?), and so the commentary and thoughts were not only entertaining but also very informative to all. I was a spectator at each of the different venues, and while we may have covered several subjects multiple times, the glaring consistency throughout is surely the need for any managers to understand their chosen risk model and its nuances if they are to benefit fully from it.
I would like to recognize the efforts of all the speakers, especially Steve, Jason, Nick, and Laurence who managed all six cities in 12 days and extend thanks to all of the 300+ attendees. Anyone interested in seeing the panel reconvened should look out for information for London in June 2012.