In this article, Rohan Douglas, the founder and CEO of Quantifi, talks about the effects of the summer’s volatility on the structured finance market, in particular how existing models have fared against the turmoil.
Q: What will be the long-term consequences of the crisis on confidence and liquidity? What kinds of institutions will have to make the biggest changes to their processes and will some of the more exotic structured finance products simply not get done any more?
A: Looking at similar crises in the past, I would expect the market to learn some lessons but bounce back even stronger than before. The structured finance and credit derivatives markets have, over time, proven to satisfy fundamental investor demands and should be expected to continue to grow steadily. There may, however, be an increased demand for structures that are more uniform and transparent. In a new world where credit risk is more widely disseminated, the science of understanding and managing systemic risk and liquidity will grow in importance.
Q: Have people got their modelling right, especially in terms of structured finance? There are questions over the role
of the rating agencies, but sophisticated institutional investors seem to have been caught by surprise as well.
A: There is some common confusion here worth clarifying. Rating agencies use models to rate CDOs where the rating is
based on the likelihood of receiving principal back at maturity. The models used by rating agencies for this purpose have a set of assumptions and typically use historical data. Rating agency models are not appropriate for marking or risk analysis and while sometimes conceptually similar, are effectively quite different from those that are best practice for pricing and hedging credit derivatives.
Structured finance products such as CDOs containing subprime loans have historically been buy-and-hold investments with little or no ability by the holder to model or analyse the risks in these transactions. For investors looking to mark their portfolio to market, it’s more often a case of no model rather than a model that failed.