RMS responds to AIR's attack on hurricane risk modelling

Published in: Risk, Risk Models, Cat risk

Companies: RMS, AIR Worldwide

A paper from AIR Worldwide on the modelling of Atlantic hurricane risk caused a minor storm in the catastrophe modelling world earlier this month.

In a thinly veiled attack on RMS, AIR criticised the use of medium-term projections based on sea surface temperatures (SSTs) as a misguided method to model Atlantic hurricanes.

InsuranceERM asked RMS to respond to the points raised and explain the reasons behind its chosen approach.

What is RMS's approach to modelling hurricane risk?

RMS is the only modelling company that uses a forward-looking approach in developing a medium-term view of hurricane risk. This approach projects risk based on the expected Atlantic basin conditions, beyond SSTs, over the next five years.

Other approaches purely based on a subset of history known to have warm SSTs restrict the "current" view of risk to only conditions and seasons from the past and cannot anticipate unique conditions not seen before that may occur in the future. These approaches may therefore be unable to capture the actual current and future risk to a portfolio.

RMS is also the only company to have model developers assigned full-time to executing and improving the medium-term methodology, engaging academic experts, surveying the latest peer-reviewed research, and incorporating the latest data from recent hurricanes.

This stands in stark contrast to other near-term modelling approaches and methodologies that have not been updated in several years, which risk advising the market with outdated information.

Does AIR's paper accurately characterise RMS's medium-term rate (MTR) model?

The approach described in the brief does not accurately describe the RMS medium-term forecast. RMS takes an ensemble approach to address uncertainties in predicting activity beyond a single season and includes published scientific theories, beyond the influence of SSTs, that explain the oscillation in high and low multidecadal frequency hurricane periods seen throughout history. References made to modelling companies using 'expert elicitation' are about a decade out of date.

We regularly receive feedback that the market looks to modelling companies to provide a reference view of current risk that advises on the latest state of hurricane activity and science. The medium-term rate forecast meets this market need by offering an ensemble view compiled from objective analysis and scoring against the entire historical record.

In addition, RMS offers its clients a long-term view of hurricane risk and several plausible views of medium-term risk, each based on a different set of assumptions. It also openly engages its licensing clients in regular discussion about the science and assumptions underlying those views of risk. This allows RMS' clients to make informed decisions when choosing their own view of risk that is suited to the appropriate business problem being addressed.

How do you respond to AIR's assertion that "Any promise of superior results using near-term forecasting is not supported by science, and the volatility it introduces is a disservice to model users."

Following every hurricane season, including years in which the medium-term forecast is not updated, RMS performs internal reviews of the forecast with the latest available data to ensure that it remains more skilful than a long-term approach.

In comparisons of historical data to 'what-if' forecasts in which the MTR model projects activity in every sequential five-year period in the last 50 years, the MTR model has proven more skilful than the analogous long-term average in predicting observed five-year activity.

Within this 50-year period, the MTR model correctly captures known phase changes in hurricane activity driven by changes in SSTs; approaches conditioned on historical warm SST years cannot capture these phase changes.

The MTR approach is responsive to the latest scientific insights, while a view that focuses on stability alone and does not consider the most recent science may provide the industry with a very inaccurate risk projection. As demonstrated in our recent blog article, the MTR approach allows RMS to proactively engage the market about upcoming MTR updates well in advance of release to the market, reducing the impact of potential volatility.

We engage our clients frequently through detailed conversations and documentation that transparently highlight key assumptions based on the latest hurricane risk and scientific research.

Has there been any discussion within RMS about the validity of its MTR model?

As mentioned above, the annual performance of MTR skilfulness is analysed at the conclusion of each hurricane season by experts across RMS. These analyses continue to instil confidence that the MTR approach remains more skilful than long-term and other near-term approaches.

Does RMS have any plans to reform or withdraw the MTR model?

Given its skill over the long-term, RMS has no plans to withdraw the MTR model. We will continue to enhance the methodology by using the latest data and research and engaging with academia and other research organisations to bring the most up-to-date view of hurricane risk available to the market.

Given the nature of hurricane risk and the uncertainty associated with each model component, providing this alternative view of risk adds significant value for our clients.

Delivery of the RMS medium-term forecast to the insurance market for the last 12 years highlights the company's commitment to quality and transparency when engaging its clients on how to build their own view of risk.

Much has been said about a modeller's "responsibility" to the insurance market, but we believe this responsibility is to enable companies with the flexibility to make informed decisions based on their risk selection criteria and an open dialogue about underlying model assumptions. RMS meets this responsibility through its medium-term approach and will strive to continue to meet it into the future.