RaisePartner | RP Market Watch - Towards new stress-test practices in the HF industry

RP Market Watch

Choose :

(12 à 12 sur 48)

Towards new stress-test practices in the HF industry

Posted on 13-Aug-2009

The Hedge Fund industry underwent a real-life stress test in 2008: from $1.9 trillion, AUM dropped to $1.4 trillion in 2008 and then to about $1 trillion today, resulting into the liquidation of about 1 400 hedge funds last year (see [1]). What kind of stress tests could help cover these kinds of worst-case scenarios?

Clearly, most existing stress tests failed to identify potential problems during the financial crisis. For instance, the stress tests defined by the Fed are based on "what-if" macro scenarios (GDP, unemployment and house prices) and might fail to predict the impact of risk factors that are specific to an institution or a portfolio.
“Reverse stress testing” (see [2]) overcomes some of the limits of traditional stress tests and can be used as a complement. These reverse stress tests follow a “bottom-um” approach (as opposed to the “top-down” traditional stress tests based on macro scenarios) as they try to determine what specific factors could impact a given portfolio. But it might come short when brutal market changes occur.

In any stress-test approach, one of the underlying challenges is to develop a risk model which is:
- dynamic enough to detect new risk factors as they arise,
- reliable enough not to misinterpret local market noise as a real turmoil.

A good risk model should help define the worst case scenario by identifying the factors to which a specific portfolio is most sensitive. But many Hedge Fund managers have been using models that failed to take into consideration certain risk factors, simply because they had never been seen as risks before, and were not considered as such in their “what-if” scenarios.

RaisePartner provides risk models that combine unique key properties for the design of tailored stress-tests:
- Specificity is obtained by using a “full-rank” approach when building the risk model. Indeed, standard factorial approaches are broadly used to explain the “ex-post” decomposition of the risk but they might fail to detect the rise of new risk factors in an “ex-ante” risk model.
- Reactivity is achieved by using short estimation windows for the estimation of volatilities and correlations. For instance, a change of sign for a cross-sector correlation can be detected much sooner.
- Reliability is ensured by a patented risk filtering method that distinguishes noise from real market information.

Another key ingredient when defining stress-tests is high performance computing: obtaining the stress-test results in real-time on large universes with a low CPU-cost is crucial for a concrete use of these results. When computing stress-tests at the global firm level, several sectors have to be aggregated, leading to very large universes.

With appropriate risk models and high-performance computing approaches (code tuning), RaisePartner provides unique functionalities for relevant and efficient stress-test computations.